# Streaming

Streaming is the incremental delivery of data to a client in chunks as it becomes available, rather than waiting for the complete response.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## AURA: AUdio-dRiven streaming Avatar

DevFeed: [AURA: AUdio-dRiven streaming Avatar](<https://devfeed.tech/articles/aura-audio-driven-streaming-avatar-145682.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.26540>)

Author: Nathaniel Cohen, Nicolas Dufour, Am\\'elie Royer, Alasdair Newson, Patrick P\\'erez

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Streaming](<https://devfeed.tech/topics/streaming.md>), [Generative AI](<https://devfeed.tech/topics/generative-ai.md>), [Real-time AI interactions](<https://devfeed.tech/topics/real-time-ai-interactions.md>)

Tags: [attention](<https://devfeed.tech/tags/attention.md>), [audio](<https://devfeed.tech/tags/audio.md>), [aura](<https://devfeed.tech/tags/aura.md>), [avatar](<https://devfeed.tech/tags/avatar.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [cache](<https://devfeed.tech/tags/cache.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [denoising](<https://devfeed.tech/tags/denoising.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [distillation](<https://devfeed.tech/tags/distillation.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [faster](<https://devfeed.tech/tags/faster.md>), [fps](<https://devfeed.tech/tags/fps.md>), [generation](<https://devfeed.tech/tags/generation.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [interactive](<https://devfeed.tech/tags/interactive.md>), [kv-cache](<https://devfeed.tech/tags/kv-cache.md>), [latency](<https://devfeed.tech/tags/latency.md>), [streaming](<https://devfeed.tech/tags/streaming.md>)

### AI overview

AURA generates audio-driven avatar video chunk by chunk using a rolling KV cache, so generation can begin with audio from the current chunk rather than requiring multi-second look-ahead. Its 1.3B model streams 576p at 20.7 fps on one GPU for 16 Hz playback; the 5B model streams 720p at 9.7 fps. A two-stage distillation process aligns a causal student to a teacher with fewer diffusion steps, then uses Distribution Matching Distillation with backward simulation. On standard talking-head benchmarks, the 1.3B model is 23x faster than its bidirectional teacher and has quality comparable to TalkVerse at lower resolution; the 5B model matches its TalkVerse teacher at 720p while running 17x faster. The authors report releasing both models' weights, 30K synthetic videos, and precomputed teacher trajectories.

### Source excerpt

arXiv:2611.26540v1 Announce Type: cross Abstract: Current audio-driven video diffusion models can synthesize realistic talking-head videos but generally require dozens of diffusion denoising steps over multi-second video chunks, making them unsuitable for streaming applications such as voice assistants and virtual avatars, where low-latency and online causal generation are required properties. AURA generates video chunk by chunk with a rolling KV cache, requiring only the audio of the current chunk to start generating, rather than a multi-second audio look-ahead like bidirectional models. It comes in two sizes: AURA 1.3B streams 576p video in real time on a single GPU (20.7 fps for a 16 Hz playback rate), making it suited to interactive avatar applications, and AURA 5B streams 720p video at 9.7 fps. Our two-stage distillation pipeline first aligns the causal student with the teacher, by training it to match the teacher's trajectories with fewer diffusion steps and a chunk-wise sliding window causal attention mask, then applies Distribution Matching Distillation (DMD) with backward simulation to close the remaining quality gap. On standard talking-head benchmarks, AURA 1.3B runs 23x faster than its own bidirectional teacher and reaches a quality comparable to TalkVerse at a lower resolution, and AURA 5B matches its TalkVerse teacher at 720p while running 17x faster. To support reproducibility and further research on streaming avatar generation, we release the weights of both models along with a dataset of 30K synthetic videos and their precomputed teacher trajectories.

## PEARL: Personalized Streaming Video Understanding with Runtime-Defined Concepts

DevFeed: [PEARL: Personalized Streaming Video Understanding with Runtime-Defined Concepts](<https://devfeed.tech/articles/pearl-personalized-streaming-video-understanding-with-runtime-defined-concepts-142515.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2603.20422>)

Author: Yuanhong Zheng, Ruichuan An, Xiaopeng Lin, Yuxing Liu, Sihan Yang, Huanyu Zhang, Haodong Li, Qintong Zhang, Renrui Zhang, Guopeng Li, Yifan Zhang, Yuheng Li, Wentao Zhang

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Transformers](<https://devfeed.tech/topics/transformers.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>), [object-detection](<https://devfeed.tech/topics/object-detection.md>)

Tags: [actions](<https://devfeed.tech/tags/actions.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [history](<https://devfeed.tech/tags/history.md>), [streaming](<https://devfeed.tech/tags/streaming.md>), [streaming-video](<https://devfeed.tech/tags/streaming-video.md>), [user](<https://devfeed.tech/tags/user.md>), [video](<https://devfeed.tech/tags/video.md>), [vision-language-models-vlms](<https://devfeed.tech/tags/vision-language-models-vlms.md>)

### AI overview

PEARL targets personalized streaming video understanding: it lets users define concepts during a video stream and answer questions about their current state or earlier appearances. It stores concept definitions separately from archived clips, then rewrites queries with concept descriptions to retrieve relevant history. The authors introduce PSVU-Bench, with 132 videos and timestamped questions covering concept definition, real-time QA, and past-time QA, and PSVU-10K to train PEARL's retriever and answering VLM once; concepts defined at test time need no training. Across three architectures, PEARL without additional training improves real-time and past-time QA by averages of 15.05% and 12.60% over corresponding offline models, and surpasses existing online video models on both. Training further improves results on Qwen3-VL-8B by 8.66% and 9.39%, respectively.

### Source excerpt

arXiv:2603.20422v2 Announce Type: replace-cross Abstract: Unlike humans, who learn new concepts from a continuous visual stream, existing personalized vision-language models (VLMs) internalize a predefined concept set during training and operate on static images or offline videos. They cannot register a concept that a user defines in the middle of a stream, nor recall the earlier moments in which it appeared. We formalize this task as Personalized Streaming Video Understanding (PSVU) and introduce PSVU-Bench, the first benchmark for this task, which annotates 132 videos with timestamped questions. Its Concept-Definition, Real-Time and Past-Time QA ask a model to register concepts at arbitrary timestamps, recognize their current state, including user-defined actions, and retrieve their historical evidence. We propose PEARL, a strong plug-and-play baseline that stores user-defined concepts apart from archived video clips and rewrites each query with concept descriptions to retrieve the relevant history. Grounding a named concept in video is a skill shared by all concepts. We therefore construct PSVU-10K, a training set of verified concept questions on public internet videos, and train the retriever and answering VLM of PEARL on it once. Concepts defined at test time still require no training. Applied to three architectures without training, PEARL improves Real-Time QA by 15.05% and Past-Time QA by 12.60% on average over the corresponding offline models, and all three surpass existing online video models on both QA types. Training further improves PEARL on Qwen3-VL-8B by 8.66% on Real-Time QA and 9.39% on Past-Time QA, with gains in every category. Code, data and models are available at https://github.com/Yuanhong-Zheng/PEARL.

## Incremental Roseland: Batch-wise Landmark Diffusion via Kernelized Sequential Karhunen--Lo\\\`eve Update

DevFeed: [Incremental Roseland: Batch-wise Landmark Diffusion via Kernelized Sequential Karhunen--Lo\\\`eve Update](<https://devfeed.tech/articles/incremental-roseland-batch-wise-landmark-diffusion-via-kernelized-sequential-karhunen-lo-eve-update-143390.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23414>)

Author: Jeanie Schreiber

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>)

Tags: [batch](<https://devfeed.tech/tags/batch.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-na](<https://devfeed.tech/tags/cs-na.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [drift](<https://devfeed.tech/tags/drift.md>), [embedding](<https://devfeed.tech/tags/embedding.md>), [incremental](<https://devfeed.tech/tags/incremental.md>), [kernel](<https://devfeed.tech/tags/kernel.md>), [math-fa](<https://devfeed.tech/tags/math-fa.md>), [math-na](<https://devfeed.tech/tags/math-na.md>), [matrix](<https://devfeed.tech/tags/matrix.md>), [memory](<https://devfeed.tech/tags/memory.md>), [singular-value-decomposition](<https://devfeed.tech/tags/singular-value-decomposition.md>), [streams](<https://devfeed.tech/tags/streams.md>)

### AI overview

Incremental Roseland (iRoseland) updates a diffusion-style embedding as data arrive, using a kernelized Sequential Karhunen-Loève scheme to update the data-to-landmark affinity matrix's SVD. A first-order degree correction around a reference mean kernel row allows recovery of the landmark-side eigenbasis with one landmark-by-landmark eigendecomposition per step. The method makes a single pass and uses bounded memory independent of the number of processed points. A forgetting factor supports tracking drifting distributions, while a sample-size-invariant embedding keeps outputs comparable over time. The abstract reports round-off agreement with batch Roseland on stationary data and, on long streams, accuracy an order of magnitude higher than equal-memory Roseland. For drifting data, forgetting with a window matched to the drift yields large gains.

### Source excerpt

arXiv:2611.23414v1 Announce Type: new Abstract: Landmark diffusion (Roseland) computes a diffusion-map-style spectral embedding from the singular value decomposition of a degree-normalized data-to-landmark kernel matrix, at a cost far below that of full diffusion maps. It assumes, however, that all data are available at once. We introduce incremental Roseland (iRoseland), which casts the landmark affinity as a row-growing matrix whose SVD can be updated with a kernelized Sequential Karhunen--Lo\`eve (SKL) scheme. Degre correction is handled by a first order expansion about a reference mean kernel row and so that the landmark-side eigenbasis may be recovered from a single landmark-by-landmark eigendecomposition at each step. The presented method requires only a single pass over the data and runs in bounded-memory with state independent of the number of points processed. A forgetting factor lets the method track drifting distributions, and a sample-size-invariant embedding keeps outputs comparable across time. Experiments show agreement with Roseland to round-off on stationary data, and accuracy that exceeds equal-memory Roseland by an order of magnitude on long streams. On drifting data, we observe large gains from forgetting with a window matched to the drift.

## StreamPhy: Streaming Inference of High-Dimensional Physical Dynamics via State Space Models

DevFeed: [StreamPhy: Streaming Inference of High-Dimensional Physical Dynamics via State Space Models](<https://devfeed.tech/articles/streamphy-streaming-inference-of-high-dimensional-physical-dynamics-via-state-space-models-143254.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2605.07384>)

Author: Panqi Chen, Yifan Sun, Shikai Fang, Xiao Fu, Lei Cheng

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [streaming data analytics](<https://devfeed.tech/topics/streaming-data-analytics.md>), [Mamba](<https://devfeed.tech/topics/mamba.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [encoder](<https://devfeed.tech/tags/encoder.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [evolution](<https://devfeed.tech/tags/evolution.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [faster](<https://devfeed.tech/tags/faster.md>), [field](<https://devfeed.tech/tags/field.md>), [fields](<https://devfeed.tech/tags/fields.md>), [framework](<https://devfeed.tech/tags/framework.md>), [functional](<https://devfeed.tech/tags/functional.md>), [generation](<https://devfeed.tech/tags/generation.md>), [generative](<https://devfeed.tech/tags/generative.md>), [generative-models](<https://devfeed.tech/tags/generative-models.md>), [improvement](<https://devfeed.tech/tags/improvement.md>), [inference](<https://devfeed.tech/tags/inference.md>), [models](<https://devfeed.tech/tags/models.md>), [state-space](<https://devfeed.tech/tags/state-space.md>), [state-space-models](<https://devfeed.tech/tags/state-space-models.md>), [streaming](<https://devfeed.tech/tags/streaming.md>)

### AI overview

StreamPhy estimates evolving physical fields in real time from irregular sparse measurements. It combines an observation-pattern-adaptive encoder, a structured state-space model for memory-efficient updates across irregular time intervals, and an FT-FiLM decoder that generates continuous fields. The authors prove FT-FiLM can represent a richer function class than the functional Tucker model. In experiments on three physical systems with challenging sampling patterns, they report at least 48% better accuracy than state-of-the-art baselines and inference 20-100 times faster than diffusion-based methods. The available text is an abstract and does not specify the systems, evaluation setup, or comparisons behind these results.

### Source excerpt

arXiv:2605.07384v4 Announce Type: replace Abstract: Inferring the evolution of high-dimensional and multi-modal (e.g., spatio-temporal) physical fields from irregular sparse measurements in real time is a fundamental challenge in science and engineering. Existing approaches, including diffusion-based generative models and functional tensor methods, typically operate in offline settings, depend on full temporal observations, or incur substantial inference cost. We propose StreamPhy, an end-to-end framework that enables efficient and accurate streaming inference of full-field physical dynamics from incoming irregular sparse measurements. The framework integrates a data-adaptive observation encoder that is robust to arbitrary observation patterns, a structured state-space model that supports memory-efficient online updates across irregular time intervals, and an expressive Functional Tensor Feature-wise Linear Modulation (FT-FiLM) decoder for continuous-field generation. We prove that FT-FiLM is more expressive than the functional Tucker model, admitting a richer function class for handling complex dynamics. Experiments on three representative physical systems under challenging sampling patterns show that StreamPhy consistently outperforms state-of-the-art baselines, with at least 48\% improvement in accuracy and up to 20--100X faster inference than diffusion-based methods.

## High-throughput Transfer Manager for Amazon S3 in Rust is now generally available

DevFeed: [High-throughput Transfer Manager for Amazon S3 in Rust is now generally available](<https://devfeed.tech/articles/high-throughput-transfer-manager-for-amazon-s3-in-rust-is-now-generally-available-144014.md>)

Original publisher: [Read original article](<https://aws.amazon.com/blogs/developer/high-throughput-transfer-manager-for-amazon-s3-in-rust-is-now-generally-available/>)

Author: Nitin Prakash

Published: 2026-11-18T23:48:54Z

Content type: release

Language: en

Sources: [AWS Developer Tools Blog](<https://devfeed.tech/sources/aws-developer-tools-blog.md>)

Topics: [Amazon S3](<https://devfeed.tech/topics/amazon-s3.md>), [Rust](<https://devfeed.tech/topics/rust.md>), [Tokio](<https://devfeed.tech/topics/tokio.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>)

Tags: [2025](<https://devfeed.tech/tags/2025.md>), [amazon-s3](<https://devfeed.tech/tags/amazon-s3.md>), [amazon-simple-storage-service-s3](<https://devfeed.tech/tags/amazon-simple-storage-service-s3.md>), [aws-sdk](<https://devfeed.tech/tags/aws-sdk.md>), [backpressure](<https://devfeed.tech/tags/backpressure.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [behavior](<https://devfeed.tech/tags/behavior.md>), [bounded](<https://devfeed.tech/tags/bounded.md>), [buffer-pool](<https://devfeed.tech/tags/buffer-pool.md>), [caching](<https://devfeed.tech/tags/caching.md>), [cargo](<https://devfeed.tech/tags/cargo.md>), [developer-preview](<https://devfeed.tech/tags/developer-preview.md>), [developer-tools](<https://devfeed.tech/tags/developer-tools.md>), [diagnostics](<https://devfeed.tech/tags/diagnostics.md>), [directory](<https://devfeed.tech/tags/directory.md>), [dns](<https://devfeed.tech/tags/dns.md>), [download](<https://devfeed.tech/tags/download.md>), [downloads](<https://devfeed.tech/tags/downloads.md>), [engine](<https://devfeed.tech/tags/engine.md>), [launch](<https://devfeed.tech/tags/launch.md>), [library](<https://devfeed.tech/tags/library.md>), [memory](<https://devfeed.tech/tags/memory.md>), [rust](<https://devfeed.tech/tags/rust.md>), [s3](<https://devfeed.tech/tags/s3.md>), [throughput](<https://devfeed.tech/tags/throughput.md>), [uncategorized](<https://devfeed.tech/tags/uncategorized.md>)

### AI overview

The Rust Transfer Manager for Amazon S3 is generally available with a redesigned engine for high-throughput transfers. By default, it starts one worker per detected core, reuses connections across transfers, and uses a fair scheduler intended to keep large transfers from starving small ones; work can be distributed across multiple network interfaces when configured. A shared buffer pool bounds memory, defaulting to up to a quarter of available memory with a 32 GiB cap. Downloads write parts directly to their final positions, while streaming delivers parts in order with backpressure. Optional runtime diagnostics expose buffer snapshots and transfer traces through environment variables. The library also supports `wasm32-wasi*` targets. Getting started requires AWS credentials with the stated S3 permissions, a Tokio-based Rust project, and a bucket. The excerpt does not include the dependency or client setup code.

### Source excerpt

High-throughput transfers to and from Amazon Simple Storage Service (Amazon S3) often require manual tuning of threads, connections, and memory. Transfer Manager for Amazon S3, a high-level transfer library for Rust, handles this work for you. It is now generally available (GA). Since the developer preview launch in March 2025, we have redesigned the library. [...]

## Announcing the developer preview of the AWS SDK for Python v4

DevFeed: [Announcing the developer preview of the AWS SDK for Python v4](<https://devfeed.tech/articles/announcing-the-developer-preview-of-the-aws-sdk-for-python-v4-144012.md>)

Original publisher: [Read original article](<https://aws.amazon.com/blogs/developer/announcing-developer-preview-aws-sdk-for-python-v4/>)

Author: Jonathan Gaytan

Published: 2026-11-18T23:27:52Z

Content type: release

Language: en

Sources: [AWS Developer Tools Blog](<https://devfeed.tech/sources/aws-developer-tools-blog.md>)

Topics: [SDK](<https://devfeed.tech/topics/sdk.md>), [Amazon S3](<https://devfeed.tech/topics/amazon-s3.md>), [Amazon Web Services](<https://devfeed.tech/topics/aws.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>)

Tags: [amazon-bedrock](<https://devfeed.tech/tags/amazon-bedrock.md>), [amazon-dynamodb](<https://devfeed.tech/tags/amazon-dynamodb.md>), [amazon-s3](<https://devfeed.tech/tags/amazon-s3.md>), [amazon-simple-storage-service-amazon-s3](<https://devfeed.tech/tags/amazon-simple-storage-service-amazon-s3.md>), [announcements](<https://devfeed.tech/tags/announcements.md>), [async](<https://devfeed.tech/tags/async.md>), [asynchronous](<https://devfeed.tech/tags/asynchronous.md>), [aws](<https://devfeed.tech/tags/aws.md>), [aws-lambda](<https://devfeed.tech/tags/aws-lambda.md>), [aws-sdk](<https://devfeed.tech/tags/aws-sdk.md>), [aws-sdk-for-python](<https://devfeed.tech/tags/aws-sdk-for-python.md>), [aws-sdk-python](<https://devfeed.tech/tags/aws-sdk-python.md>), [boto3](<https://devfeed.tech/tags/boto3.md>), [botocore](<https://devfeed.tech/tags/botocore.md>), [clients](<https://devfeed.tech/tags/clients.md>), [developer-preview](<https://devfeed.tech/tags/developer-preview.md>), [developer-tools](<https://devfeed.tech/tags/developer-tools.md>), [dynamodb](<https://devfeed.tech/tags/dynamodb.md>), [event-streaming](<https://devfeed.tech/tags/event-streaming.md>), [modular](<https://devfeed.tech/tags/modular.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [python](<https://devfeed.tech/tags/python.md>), [s3](<https://devfeed.tech/tags/s3.md>), [sdk](<https://devfeed.tech/tags/sdk.md>), [services](<https://devfeed.tech/tags/services.md>), [streaming](<https://devfeed.tech/tags/streaming.md>)

### AI overview

AWS SDK for Python v4 is in developer preview, with typed, modular service clients and both synchronous and native async APIs. The excerpt shows installing a service package such as `aws-sdk-s3`, configuring credentials and Region, and making calls with context-managed clients; async calls can run concurrently, and async clients support event streams. Python 3.12 or later, an AWS account, and configured credentials and Region are required. Packages can supply additional credential sources, while bidirectional streaming requires the AWS Common Runtime HTTP/2 client. S3 support includes streaming transfers and flexible checksums. APIs may change before general availability: pin package versions, evaluate in a test environment, and continue using Boto3 for production. Planned additions include more services and credential providers, S3 Transfer Manager, paginators, and waiters.

### Source excerpt

The AWS SDK for Python v4 is now available in developer preview. It provides typed, modular clients for a growing list of AWS services, including Amazon Simple Storage Service (Amazon S3), Amazon DynamoDB, AWS Lambda, and Amazon Bedrock. Clients are available in synchronous and asynchronous (async) versions. Some features, such as event streaming, are currently available [...]

## A modern dilemma: Consume data from a stream or a lake

DevFeed: [A modern dilemma: Consume data from a stream or a lake](<https://devfeed.tech/articles/a-modern-dilemma-consume-data-from-a-stream-or-a-lake-141161.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/a-modern-dilemma-consume-data-from-a-stream-or-a-lake-296bbb354094?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-11-18T17:35:14Z

Content type: article

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [software-architecture](<https://devfeed.tech/topics/software-architecture.md>), [stream-processing](<https://devfeed.tech/topics/stream-processing.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>), [etl](<https://devfeed.tech/topics/etl.md>), [deltalake](<https://devfeed.tech/topics/deltalake.md>), [Apache Kafka](<https://devfeed.tech/topics/apache-kafka.md>), [Apache Flink](<https://devfeed.tech/topics/apache-flink.md>), [parquet](<https://devfeed.tech/topics/parquet.md>), [Data Quality](<https://devfeed.tech/topics/data-quality.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [apis](<https://devfeed.tech/tags/apis.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [batch](<https://devfeed.tech/tags/batch.md>), [big-data](<https://devfeed.tech/tags/big-data.md>), [business-processes](<https://devfeed.tech/tags/business-processes.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [cloud-computing](<https://devfeed.tech/tags/cloud-computing.md>), [data](<https://devfeed.tech/tags/data.md>), [data-engineering](<https://devfeed.tech/tags/data-engineering.md>), [data-lakes](<https://devfeed.tech/tags/data-lakes.md>), [delta-lake](<https://devfeed.tech/tags/delta-lake.md>), [enrichment](<https://devfeed.tech/tags/enrichment.md>), [etl](<https://devfeed.tech/tags/etl.md>), [event](<https://devfeed.tech/tags/event.md>), [flink](<https://devfeed.tech/tags/flink.md>), [kafka](<https://devfeed.tech/tags/kafka.md>), [latency](<https://devfeed.tech/tags/latency.md>), [legacy](<https://devfeed.tech/tags/legacy.md>), [log](<https://devfeed.tech/tags/log.md>), [mapping](<https://devfeed.tech/tags/mapping.md>), [model-context-protocol](<https://devfeed.tech/tags/model-context-protocol.md>), [software-architecture](<https://devfeed.tech/tags/software-architecture.md>), [software-engineering](<https://devfeed.tech/tags/software-engineering.md>), [stream](<https://devfeed.tech/tags/stream.md>)

### AI overview

Choosing a data consumption pattern early can avoid costly architectural changes later. APIs support synchronous, record-by-record requests and paginated multirecord results, making them suitable for low-latency needs; Model Context Protocol can provide similar access for agents and is often an API wrapper. Event streams, commonly built with Kafka, Flink or Spark Streaming, suit sub-second latency and enable per-event calculations, transformations and enrichment. The excerpt also identifies Delta Change Data Feed (CDF) as a record-by-record pattern and traditional batch ETL as a multirecord pattern, but ends before explaining CDF's mechanics or comparing these approaches further.

### Source excerpt

Data is a critical component of software architecture, providing the foundation for business processes, including those powered by artificial intelligence (AI). Quality, volume, latency, consistency, ordering and duplicates are essential considerations in software design. Modern cloud architectures consume data in various ways, such as through application programming interfaces (APIs), streams and data lakes. You can consume data at the individual event or record level as it arrives or in larger batches. Legacy architectures often process data in scheduled batches, forcing surrounding architecture and business processes to depend on that approach. Getting the data consumption approach right up front is vital, because rearchitecting later is costly. General patterns to consume data There are several general data consumption patterns. These include APIs (record-by-record or multirecord results), event stream consumption (record-by-record), Delta Change Data Feed (CDF) consumption (record-by-record) or traditional batch ETL-extract, transform, load (multirecord results). APIs are ideal for low-latency, record-by-record requirements. They also support multirecord results, typically through pagination. APIs usually follow a synchronous approach, where the client makes a call and waits for a response before moving to the next step. Model Context Protocol is an industry specification for agents to consume data and can support record-by-record or batch results via pagination, similar to APIs. They are often a wrapper on top of APIs. The event stream pattern works well for sub-second latency requirements and typically runs on a tech stack like Kafka, Flink or Spark Streaming. These technologies excel at record-level calculations (like on-the-fly calculation), transformations (like column mapping) or enrichment (like adding data to an existing event). Delta Lake is often used in data lakes and serves as a transaction log to standard Apache Parquet data files (source). CDF is

## datasette-arrows 0.1a0

DevFeed: [datasette-arrows 0.1a0](<https://devfeed.tech/articles/datasette-arrows-0-1a0-142012.md>)

Original publisher: [Read original article](<https://simonwillison.net/2026/Nov/18/datasette-arrows/>)

Author: Simon Willison

Published: 2026-11-18T17:09:59Z

Content type: release

Language: en

Sources: [Simon Willison's Weblog](<https://devfeed.tech/sources/simon-willison-s-weblog.md>)

Topics: [apache-arrow](<https://devfeed.tech/topics/apache-arrow.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>), [CSV](<https://devfeed.tech/topics/csv.md>)

Tags: [apache-arrow](<https://devfeed.tech/tags/apache-arrow.md>), [csv](<https://devfeed.tech/tags/csv.md>), [datasette](<https://devfeed.tech/tags/datasette.md>), [datasette-1-568](<https://devfeed.tech/tags/datasette-1-568.md>), [file-formats](<https://devfeed.tech/tags/file-formats.md>), [plugins](<https://devfeed.tech/tags/plugins.md>), [projects](<https://devfeed.tech/tags/projects.md>), [projects-567](<https://devfeed.tech/tags/projects-567.md>), [streaming](<https://devfeed.tech/tags/streaming.md>), [tsv](<https://devfeed.tech/tags/tsv.md>)

### AI overview

Datasette 1.0a43 lets plugins implement streaming export formats through the `register_output_renderer()` hook. The `datasette-arrows` plugin applies this to Apache Arrow, allowing exports to stream in chunks instead of loading all rows into memory, as Datasette already does for CSV and TSV. Install it with `datasette install datasette-arrows` to add an Arrow option to the export dialog. In the version described, Arrow is not yet grouped with CSV and TSV under "All rows in this table"; the excerpt says that grouping is planned for a later alpha release.

### Source excerpt

Release: datasette-arrows 0.1a0 Datasette 1.0a43 added the ability for plugins to implement streaming export formats. Datasette core supports streaming CSV and TSV - you can export millions of lines of data from a table without needing to load it all into memory, because those two file formats support streaming data in chunks. The register_output_renderer() plugin hook can now be used to implement the same pattern for other formats. I've learned never to release a new plugin feature without accompanying it with a plugin that uses it. Claude Fable 5.2 suggested the Apache Arrow streaming format as a good option, and built this plugin in almost a single shot. Add the plugin to Datasette (datasette install datasette-arrows) and the export dialog gains this new Arrow option: This screenshot is from Datasette's main branch - the most recent release doesn't yet group streaming formats such as Arrow with the TSV and CSV in the "All rows in this table" section, but that's coming in the next alpha release. Tags: datasette, projects

## Streaming On-Policy Distillation: Decoupling Three Clocks for Stage-Wise Supervision Consumption

DevFeed: [Streaming On-Policy Distillation: Decoupling Three Clocks for Stage-Wise Supervision Consumption](<https://devfeed.tech/articles/streaming-on-policy-distillation-decoupling-three-clocks-for-stage-wise-supervision-consumption-140007.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.22298>)

Author: Yun Zhu Da Wang

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>)

Tags: [budget](<https://devfeed.tech/tags/budget.md>), [checkpoint](<https://devfeed.tech/tags/checkpoint.md>), [clocks](<https://devfeed.tech/tags/clocks.md>), [completion](<https://devfeed.tech/tags/completion.md>), [computing](<https://devfeed.tech/tags/computing.md>), [continuation](<https://devfeed.tech/tags/continuation.md>), [control](<https://devfeed.tech/tags/control.md>), [cost](<https://devfeed.tech/tags/cost.md>), [coverage](<https://devfeed.tech/tags/coverage.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [data](<https://devfeed.tech/tags/data.md>), [decoupling](<https://devfeed.tech/tags/decoupling.md>), [delivery](<https://devfeed.tech/tags/delivery.md>), [distillation](<https://devfeed.tech/tags/distillation.md>), [distilling](<https://devfeed.tech/tags/distilling.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [gap](<https://devfeed.tech/tags/gap.md>), [policy](<https://devfeed.tech/tags/policy.md>), [stage](<https://devfeed.tech/tags/stage.md>), [streaming](<https://devfeed.tech/tags/streaming.md>), [supervision](<https://devfeed.tech/tags/supervision.md>)

### AI overview

Streaming on-policy distillation (OPD) delivers newly generated response segments while generation continues, preserving each segment's full causal prefix but computing loss only on new positions. A consumption protocol manages supervision budgets, position selection, staleness, and update timing across training stages. In GSM8K experiments distilling Qwen3-4B into Qwen3-0.6B with VERL, budget matching narrowed an early test-accuracy gap across seven common candidates from 4.55 to 2.42 points. Synchronous continuation reached 73.39% best accuracy, below synchronous training's 74.30% and above the budget-matched run's 71.95%; a 76.12% extended record was unpaired. Freshness filtering and disagreement-overlap selection improved local peaks without showing sustained advantages. A prototype reduced data-supply waiting time but did not establish equal-update speedup; end-to-end gains must account for continuation cost.

### Source excerpt

arXiv:2611.22298v1 Announce Type: new Abstract: In on-policy distillation (OPD), response completion, supervision delivery, and parameter updates need not share the same granularity. We develop streaming OPD by decoupling these three clocks: a response repeatedly delivers newly generated segments while generation continues, retaining the full causal prefix at each delivery and computing loss only on new positions. A consumption protocol then organizes supervision budgets, position selection, staleness handling, and update scheduling, allowing control over quantity, coverage, and version lag across training stages. In GSM8K experiments distilling Qwen3-4B into Qwen3-0.6B with VERL, supervised-budget matching reduces the early gap between best test accuracies over seven common candidates from 4.55 to 2.42 points. Synchronous continuation from the same checkpoint reaches 73.39% best test accuracy over seven common candidates at u=384,...,576, versus 74.30% synchronous and 71.95% budget-matched (differences -0.91 and -2.35 points); an unpaired extended record reaches 76.12%. Freshness filtering and disagreement-overlap selection improve local peaks but do not establish sustained advantages. A system prototype shows lower data-supply waiting times without establishing an equal-update speedup. Quality recovery and supply-side benefits are assessed separately; end-to-end gains must account for continuation cost.

## Amazon MSK Express brokers available on AWS Graviton4-based M8g instances

DevFeed: [Amazon MSK Express brokers available on AWS Graviton4-based M8g instances](<https://devfeed.tech/articles/amazon-msk-express-brokers-available-on-aws-graviton4-based-m8g-instances-140165.md>)

Original publisher: [Read original article](<https://aws.amazon.com/blogs/big-data/amazon-msk-express-brokers-available-on-aws-graviton4-based-m8g-instances/>)

Author: Charishma Makineni

Published: 2026-11-17T22:07:48Z

Content type: release

Language: en

Sources: [AWS Big Data Blog](<https://devfeed.tech/sources/aws-big-data-blog.md>)

Topics: [Amazon MSK](<https://devfeed.tech/topics/amazon-msk.md>), [Graviton](<https://devfeed.tech/topics/graviton.md>), [Amazon EC2](<https://devfeed.tech/topics/amazon-ec2.md>), [Amazon Web Services](<https://devfeed.tech/topics/aws.md>), [Amazon CloudWatch](<https://devfeed.tech/topics/amazon-cloudwatch.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [3](<https://devfeed.tech/tags/3.md>), [amazon](<https://devfeed.tech/tags/amazon.md>), [amazon-managed-streaming-for-apache-kafka-amazon-msk](<https://devfeed.tech/tags/amazon-managed-streaming-for-apache-kafka-amazon-msk.md>), [amazon-msk](<https://devfeed.tech/tags/amazon-msk.md>), [announcements](<https://devfeed.tech/tags/announcements.md>), [aws](<https://devfeed.tech/tags/aws.md>), [based](<https://devfeed.tech/tags/based.md>), [broker](<https://devfeed.tech/tags/broker.md>), [client](<https://devfeed.tech/tags/client.md>), [consumer](<https://devfeed.tech/tags/consumer.md>), [cross](<https://devfeed.tech/tags/cross.md>), [express](<https://devfeed.tech/tags/express.md>), [faster](<https://devfeed.tech/tags/faster.md>), [foundational-100](<https://devfeed.tech/tags/foundational-100.md>), [graviton](<https://devfeed.tech/tags/graviton.md>), [graviton4](<https://devfeed.tech/tags/graviton4.md>), [latency](<https://devfeed.tech/tags/latency.md>), [p99](<https://devfeed.tech/tags/p99.md>), [pricing](<https://devfeed.tech/tags/pricing.md>), [recovery](<https://devfeed.tech/tags/recovery.md>), [size](<https://devfeed.tech/tags/size.md>), [traffic](<https://devfeed.tech/tags/traffic.md>), [with](<https://devfeed.tech/tags/with.md>)

### AI overview

Amazon MSK Express brokers on Graviton4-based M8g instances combine lower costs with better results than M7g in the described tests: broker-hour pricing is 15% lower, broker-side cross-AZ client traffic is uncharged (cutting total cross-AZ transfer cost by 50%), and P99 produce latency was 21-54% lower at the same tested load. Brokers used 9-11 percentage points less CPU, and consumers cleared up to three times more backlog in 30 minutes. Results depend on workload settings; the catch-up test used large brokers only. M8g adds a 24xlarge option, while shared sizes retain the same per-broker storage and network throughput. Teams can create an M8g cluster, update an M7g Express cluster in place one broker at a time, or migrate from Standard or self-managed Kafka using MSK Replicator. Rack-aware consumers can reduce cross-AZ charges further, though reading from follower replicas may add some end-to-end latency.

### Source excerpt

Amazon MSK Express brokers are now available on AWS Graviton4-based M8g instances. Compared with Express on M7g, M8g delivers 15% lower broker-hour pricing, lower P99 produce latency, faster consumer recovery, a new 24xlarge size, and no broker-side charge for cross-AZ client traffic.

## Qwik 2.0 RC: Replaces HTML Comment Markers for Faster Reactivity and Out-of-Order Streaming

DevFeed: [Qwik 2.0 RC: Replaces HTML Comment Markers for Faster Reactivity and Out-of-Order Streaming](<https://devfeed.tech/articles/qwik-2-0-rc-replaces-html-comment-markers-for-faster-reactivity-and-out-of-order-streaming-137155.md>)

Original publisher: [Read original article](<https://www.infoq.com/news/2026/11/quik-2-streaming/>)

Author: Daniel Curtis

Published: 2026-11-17T12:25:00Z

Content type: news

Language: en

Sources: [InfoQ](<https://devfeed.tech/sources/infoq.md>)

Topics: [Qwik](<https://devfeed.tech/topics/qwik.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>), [HTML](<https://devfeed.tech/topics/html.md>), [Web Development](<https://devfeed.tech/topics/web-development.md>), [performance-optimization](<https://devfeed.tech/topics/performance-optimization.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [async](<https://devfeed.tech/tags/async.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [cache](<https://devfeed.tech/tags/cache.md>), [candidate](<https://devfeed.tech/tags/candidate.md>), [command](<https://devfeed.tech/tags/command.md>), [development](<https://devfeed.tech/tags/development.md>), [experimental](<https://devfeed.tech/tags/experimental.md>), [html](<https://devfeed.tech/tags/html.md>), [javascript](<https://devfeed.tech/tags/javascript.md>), [migration](<https://devfeed.tech/tags/migration.md>), [news](<https://devfeed.tech/tags/news.md>), [quik-2-streaming](<https://devfeed.tech/tags/quik-2-streaming.md>), [qwik](<https://devfeed.tech/tags/qwik.md>), [route](<https://devfeed.tech/tags/route.md>), [script](<https://devfeed.tech/tags/script.md>), [state](<https://devfeed.tech/tags/state.md>), [streaming](<https://devfeed.tech/tags/streaming.md>), [typescript](<https://devfeed.tech/tags/typescript.md>), [vanilla-javascript](<https://devfeed.tech/tags/vanilla-javascript.md>), [web-development](<https://devfeed.tech/tags/web-development.md>)

### AI overview

Qwik 2.0 RC replaces v1's HTML comment markers with two script blocks for resumable state and virtual node data, removing the ordered-write constraint that prevented out-of-order streaming. In Qwik's internal js-framework-benchmark runs, v2 measured 1.23x slower than vanilla JavaScript, compared with 2.59x for v1. Experimental streaming lets `<Pending>` and `<Catch>` boundaries send the rest of a page before slow content, which arrives later in the same response; it requires opt-in Vite configuration and an inline script to swap in the content. Async `useComputed$` replaces `<Resource>`, while route loaders gain individually cacheable JSON endpoints and hover prefetching. A migration command handles the breaking changes and package renames. One walkthrough also reported less JavaScript needed before interactivity (59KB to 9KB), though total bundle size grew.

### Source excerpt

Qwik 2.0 Release Candidate, published on 1 October 2026, replaces v1's HTML comment markers with two script blocks that hold state and virtual node data. It adds experimental out-of-order streaming, async useComputed$, and cacheable route loaders. Qwik's benchmarks show v2 at 1.23x vanilla JavaScript, versus 2.59x for v1. A migration command is provided. By Daniel Curtis

## Surprise-Gated Memory for Live Speaker Diarization

DevFeed: [Surprise-Gated Memory for Live Speaker Diarization](<https://devfeed.tech/articles/surprise-gated-memory-for-live-speaker-diarization-137617.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.20819>)

Author: Louis Mouchon, Alexis Lainault

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>)

Tags: [audio](<https://devfeed.tech/tags/audio.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-sd](<https://devfeed.tech/tags/cs-sd.md>), [diarization](<https://devfeed.tech/tags/diarization.md>), [gate](<https://devfeed.tech/tags/gate.md>), [invariant](<https://devfeed.tech/tags/invariant.md>), [live](<https://devfeed.tech/tags/live.md>), [memory](<https://devfeed.tech/tags/memory.md>), [online](<https://devfeed.tech/tags/online.md>), [speaker](<https://devfeed.tech/tags/speaker.md>), [speaker-diarization](<https://devfeed.tech/tags/speaker-diarization.md>), [voice](<https://devfeed.tech/tags/voice.md>)

### AI overview

SGM-Diar applies a frozen, fine-tuned DiariZen encoder and calibrated novelty gate to live speaker diarization, storing one-shot speaker memories and consolidating them through periodic replay. In end-of-meeting revisable mode, it recovered 52% of the online-to-offline diarization-error gap on AMI-SDM and 82% on VoxConverse; calibration was frozen on development data and transferred across domains without adjustment. The reported gate had 0.857 one-shot speaker recall and a 0.001 false-novelty rate, with novelty responses habituating as speakers became familiar and re-identification unaffected by intervening silence. Two ablations found sliding-window replay worse than no replay. Results are single-seed runs on three AMI meetings and five VoxConverse sessions, so the authors emphasize condition ordering over absolute rates. Memory is limited to each meeting and erased afterward; the runtime system receives no gradient. The work is presented as a proof of concept.

### Source excerpt

arXiv:2611.20819v1 Announce Type: cross Abstract: We transpose the Surprise-Gated Memory recipe, a frozen encoder plus a calibrated novelty signal that gates a one-shot episodic store consolidated by periodic replay, from continual visual learning to live speaker diarization. The moment-to-moment question a streaming diarizer must answer, whether an arriving voice belongs to an already-heard speaker or a new one, is a known-versus-novel decision, exactly the decision the surprise gate is built for. On a frozen fine-tuned DiariZen backbone, our online system SGM-Diar recovers 4.8 of the 9.2 diarization-error points separating a naive online baseline from the offline oracle on AMI-SDM (52%) and 1.9 of 2.3 points on VoxConverse (82%), in the end-of-meeting revisable mode, with calibration frozen on development data and transferred cross-domain without adjustment. These are single-seed runs on three AMI meetings and five VoxConverse sessions, so the ordering of conditions rather than the absolute rates is the result. Beyond error rate, we measure the gate's behaviour directly and find a coherent memory-like profile: one-shot speaker recall at 0.857, a novelty response that habituates as speakers become familiar, re-identification that is invariant to the length of intervening silence, and a false-novelty rate of 0.001. Two ablations reproduce, now on audio, the central prediction of the visual SGM work: a sliding replay window is worse than no replay at all. The memory is strictly intra-meeting and is wiped at the end, and no gradient touches the run-time system. We report SGM-Diar as a proof-of-concept and state its limitations explicitly.

## STORMGUIDE-FL: Stochastic Remote Guidance for Efficient Mining of Federated Streaming Data

DevFeed: [STORMGUIDE-FL: Stochastic Remote Guidance for Efficient Mining of Federated Streaming Data](<https://devfeed.tech/articles/stormguide-fl-stochastic-remote-guidance-for-efficient-mining-of-federated-streaming-data-136995.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.20975>)

Author: Manuel R\\"oder, Frank-Michael Schleif

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Databases](<https://devfeed.tech/sources/arxiv-databases.md>), [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Federated Learning](<https://devfeed.tech/topics/federated-learning.md>), [Data Mining & Modeling](<https://devfeed.tech/topics/data-mining-modeling.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>)

Tags: [acquisition](<https://devfeed.tech/tags/acquisition.md>), [asynchronous](<https://devfeed.tech/tags/asynchronous.md>), [bounded](<https://devfeed.tech/tags/bounded.md>), [budget](<https://devfeed.tech/tags/budget.md>), [client](<https://devfeed.tech/tags/client.md>), [clients](<https://devfeed.tech/tags/clients.md>), [compute](<https://devfeed.tech/tags/compute.md>), [concept-drift](<https://devfeed.tech/tags/concept-drift.md>), [continuous](<https://devfeed.tech/tags/continuous.md>), [contributions](<https://devfeed.tech/tags/contributions.md>), [convergence](<https://devfeed.tech/tags/convergence.md>), [cs-db](<https://devfeed.tech/tags/cs-db.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [data-mining](<https://devfeed.tech/tags/data-mining.md>), [data-privacy](<https://devfeed.tech/tags/data-privacy.md>), [devices](<https://devfeed.tech/tags/devices.md>), [federated](<https://devfeed.tech/tags/federated.md>), [guidance](<https://devfeed.tech/tags/guidance.md>), [mining](<https://devfeed.tech/tags/mining.md>), [remote](<https://devfeed.tech/tags/remote.md>), [streaming](<https://devfeed.tech/tags/streaming.md>), [streaming-data](<https://devfeed.tech/tags/streaming-data.md>)

### AI overview

STORMGUIDE-FL targets local concept drift in each client's streaming data and global client drift caused by asynchronous participation and imbalanced contributions. It combines incremental client-side sampling of informative, diverse instances under bounded memory, compute and labeling budgets with server guidance applied as an update-level correction during local optimization. The correction aims to counter client drift without discarding scarce labeled samples. The abstract reports evaluations across gradual, sudden and recurring drift, with sampling gains compounding with the correction and improved speed and robustness of global convergence under resource-constrained, heterogeneous conditions.

### Source excerpt

arXiv:2611.20975v1 Announce Type: cross Abstract: Federated Learning enables collaborative model training across distributed edge devices ("clients") while preserving data privacy. However, real-world deployments involve clients that observe continuous data streams from evolving environments while operating under strict resource limitations. Existing solutions still face two major challenges: local concept drift, where each client's data distribution shifts over time, and global client drift, arising from asynchronous engagement and imbalanced contributions across heterogeneous training participants, formally described as dual client heterogeneity. We propose STORMGUIDE-FL, a federated optimization and data-mining framework that mitigates both types of drift using a server-guided pipeline for data acquisition and local training. For data acquisition, a lightweight client-side sampling strategy incrementally selects informative and diverse instances from non-stationary, non-IID local streams under bounded memory, compute, and labeling budget. For local training, the server's remote guidance is embedded directly into local optimization as an update-level correction, counteracting client drift without discarding scarce labeled samples. We empirically verify the effectiveness of the proposed sampling strategy on federated data streams across gradual, sudden, and recurring concept drift, and demonstrate that its gains compound with the update-level drift correction. Experimental results further show improved speed and robustness of global model convergence under resource-constrained, heterogeneous conditions.

## Decode Early, Verify Late: Streaming Offline Speech Recognizers Without Accuracy Loss

DevFeed: [Decode Early, Verify Late: Streaming Offline Speech Recognizers Without Accuracy Loss](<https://devfeed.tech/articles/decode-early-verify-late-streaming-offline-speech-recognizers-without-accuracy-loss-138984.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.20064>)

Author: Doyeob Kim

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [asr](<https://devfeed.tech/topics/asr.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>), [Real-time AI interactions](<https://devfeed.tech/topics/real-time-ai-interactions.md>)

Tags: [1-5](<https://devfeed.tech/tags/1-5.md>), [22](<https://devfeed.tech/tags/22.md>), [a100](<https://devfeed.tech/tags/a100.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [after](<https://devfeed.tech/tags/after.md>), [agents](<https://devfeed.tech/tags/agents.md>), [arithmetic](<https://devfeed.tech/tags/arithmetic.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [asr](<https://devfeed.tech/tags/asr.md>), [audio](<https://devfeed.tech/tags/audio.md>), [comparison](<https://devfeed.tech/tags/comparison.md>), [complete](<https://devfeed.tech/tags/complete.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [decode](<https://devfeed.tech/tags/decode.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [equals](<https://devfeed.tech/tags/equals.md>), [falls](<https://devfeed.tech/tags/falls.md>), [offline](<https://devfeed.tech/tags/offline.md>), [speech](<https://devfeed.tech/tags/speech.md>), [streaming](<https://devfeed.tech/tags/streaming.md>), [verify](<https://devfeed.tech/tags/verify.md>)

### AI overview

Decode Early, Verify Late (DEVL) streams offline ASR by showing partial transcripts as drafts, then verifying the latest draft against the complete audio in one forward pass when the turn ends. Under stated input and decoding conditions, its final output equals greedy offline decoding in exact arithmetic. Across five recognizers, 95-100% of finals were byte-identical to offline transcripts, with word error rate within 0.13 points; single-stream time to final fell 1.5-7.3x. For Qwen3-ASR-1.7B on one A100, median final latency was 22-26 ms after turn end at up to 64 streams, 2.7-3.9x sooner than its official streaming mode. The tested system met a 100 ms p99 target at up to 80 streams on Pipecat-STT and 64 on WildASR.

### Source excerpt

arXiv:2611.20064v1 Announce Type: new Abstract: Voice agents need accurate transcripts soon after a user stops speaking. We propose Decode Early, Verify Late (DEVL), a training-free framework for streaming inference with offline ASR models. The server keeps partial transcripts as drafts and verifies the latest against the complete audio in one forward pass at turn end. Under stated input and decoding conditions, the final equals the model's greedy offline output in exact arithmetic. We adapt DEVL to four model families. In our five-recognizer comparison, 95-100% of final transcripts are byte-identical to the offline transcripts, WER stays within 0.13 points of offline, and single-stream time to final falls 1.5-7.3x. Serving Qwen3-ASR-1.7B on one A100, DEVL returns finals a median 22-26 ms after turn end at up to 64 streams, 2.7-3.9x sooner than the official streaming mode. It is the only system we tested that meets a 100 ms p99 target, up to 80 streams on Pipecat-STT and 64 on WildASR.

## AdaModalX: Synchronization-Free MFM Inference for Distributed Multimodal Mobile Sensing via Progressive Data Orchestration

DevFeed: [AdaModalX: Synchronization-Free MFM Inference for Distributed Multimodal Mobile Sensing via Progressive Data Orchestration](<https://devfeed.tech/articles/adamodalx-synchronization-free-mfm-inference-for-distributed-multimodal-mobile-sensing-via-progressive-data-orchestration-139057.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.20464>)

Author: Xiangrui Xu, Sicong Liu, Zhiwen Yu, Zimu Zhou, Bin Guo

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [AI Architecture](<https://devfeed.tech/topics/ai-architecture.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>), [On-device AI](<https://devfeed.tech/topics/on-device-ai.md>)

Tags: [activation](<https://devfeed.tech/tags/activation.md>), [applications](<https://devfeed.tech/tags/applications.md>), [candidate](<https://devfeed.tech/tags/candidate.md>), [completeness](<https://devfeed.tech/tags/completeness.md>), [contribution](<https://devfeed.tech/tags/contribution.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [data](<https://devfeed.tech/tags/data.md>), [data-orchestration](<https://devfeed.tech/tags/data-orchestration.md>), [device](<https://devfeed.tech/tags/device.md>), [distributed](<https://devfeed.tech/tags/distributed.md>), [edge](<https://devfeed.tech/tags/edge.md>), [execution](<https://devfeed.tech/tags/execution.md>), [exit](<https://devfeed.tech/tags/exit.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [features](<https://devfeed.tech/tags/features.md>), [inference](<https://devfeed.tech/tags/inference.md>), [mobile](<https://devfeed.tech/tags/mobile.md>), [multimodal](<https://devfeed.tech/tags/multimodal.md>), [orchestration](<https://devfeed.tech/tags/orchestration.md>), [sensing](<https://devfeed.tech/tags/sensing.md>), [synchronization](<https://devfeed.tech/tags/synchronization.md>)

### AI overview

AdaModalX aims to reduce latency in distributed mobile sensing without waiting for every modality to arrive. A lightweight device-side probe lets reliable samples exit locally while uncertain ones continue uploading. At the edge, the system forms candidates from received data and, when needed, imputed features for delayed inputs; a contribution assessor using frozen MFM encoders estimates their utility and triggers the MFM when a candidate meets its threshold. Across four mobile sensing applications, the authors report up to 26.97% higher accuracy and up to 64.75% lower end-to-end latency. These are reported maximum gains; the excerpt does not specify baselines or per-application results.

### Source excerpt

arXiv:2611.20464v1 Announce Type: new Abstract: Multimodal foundation models (MFMs) support generalizable inference in distributed mobile sensing applications by integrating heterogeneous data streams. However, existing MFM inference pipelines typically require complete and synchronized multimodal inputs, whereas mobile sensing streams arrive asynchronously due to network variability and heterogeneous data volumes across modalities. Waiting for all modalities increases end-to-end latency, while inference with incomplete inputs can reduce accuracy. We argue that multimodal inference should depend on the utility of available information rather than modality completeness. Based on this principle, we propose AdaModalX, a progressive multimodal inference framework that decouples inference execution from modality synchronization through runtime utility-aware orchestration. At each device, importance-aware data uploading uses a lightweight probe distilled from the edge-side MFM to allow reliable samples to exit locally while uncertain samples continue uploading. At the edge, on-demand MFM activation constructs candidates from fully received samples and, when needed, imputed features derived from partially received delayed inputs. A lightweight contribution assessor built on frozen MFM encoders estimates candidate utility and activates the MFM once the candidate meets its corresponding utility threshold. Experiments on four mobile sensing applications show that AdaModalX improves the accuracy-latency trade-off, achieving up to 26.97% accuracy improvement and up to 64.75% end-to-end latency reduction.

## RESYNC: Residual Latent-State Synchronization for Deadline-Bound Streaming Inference

DevFeed: [RESYNC: Residual Latent-State Synchronization for Deadline-Bound Streaming Inference](<https://devfeed.tech/articles/resync-residual-latent-state-synchronization-for-deadline-bound-streaming-inference-132251.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.16523>)

Author: Eljas Linna, Hao Dong, Juho Kanniainen

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Streaming](<https://devfeed.tech/topics/streaming.md>), [synchronization](<https://devfeed.tech/topics/synchronization.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [all](<https://devfeed.tech/tags/all.md>), [anchor](<https://devfeed.tech/tags/anchor.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [background](<https://devfeed.tech/tags/background.md>), [beats](<https://devfeed.tech/tags/beats.md>), [best](<https://devfeed.tech/tags/best.md>), [better](<https://devfeed.tech/tags/better.md>), [but](<https://devfeed.tech/tags/but.md>), [complete](<https://devfeed.tech/tags/complete.md>), [compute](<https://devfeed.tech/tags/compute.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [fast](<https://devfeed.tech/tags/fast.md>), [faster](<https://devfeed.tech/tags/faster.md>), [inference](<https://devfeed.tech/tags/inference.md>), [state](<https://devfeed.tech/tags/state.md>), [streaming](<https://devfeed.tech/tags/streaming.md>), [synchronization](<https://devfeed.tech/tags/synchronization.md>)

### AI overview

RESYNC lets a large model compute asynchronously while a fast updater advances its latest exact but stale latent state with newly arrived observations and reuses the model's unchanged prediction head. The updater learns from the frozen model's representations without task labels. Across one real and two synthetic streams and five architecture families, it beat the strongest compact or distilled alternative in all 12 tested dataset-architecture pairs at anchor ages of 4-12 observations, and in 11 pairs at older tested ages up to 32, when the teacher had a material advantage. It preserved the teacher's gain completely on the real stream; preservation was lower and generally declined with anchor age on synthetic streams. In 1 ms live replay, synchronous large models missed every deadline, while RESYNC met at least 99.4% of arrivals with 0.35-0.62 ms median latency on one CPU thread--about as fast as or faster than the compact model, and 7.9-17.7 times faster than synchronous large-model inference. The approach targets settings with background compute but strict per-observation deadlines.

### Source excerpt

arXiv:2611.16523v1 Announce Type: cross Abstract: Predicting every observation in a high-rate stream requires a model to finish before the next arrival. A large model may offer better predictions but miss that deadline. We propose RESYNC: the large model runs asynchronously, while a fast path updates its latest exact but stale latent state using the observations received since that state was computed and applies its unchanged prediction head. The updater learns from the frozen model's representations without task labels. Across one real and two synthetic streams and five architecture families, RESYNC beats the best compact or distilled alternative with a paired 95% interval above zero in all twelve dataset--architecture pairs with a material teacher advantage at anchor ages of 4--12 observations, and in eleven pairs at older tested ages up to 32. The share of the teacher's gain preserved remains complete on the real stream but is lower and generally declines with anchor age on the synthetic streams. In live replay with 1 ms arrivals, the synchronous large models cannot deliver any predictions on time while RESYNC serves at least 99.4% of arrivals within the interval, with 0.35--0.62 ms median latency on one CPU thread: around the same speed as or faster than the compact model, and 7.9--17.7$\times$ faster than synchronous large-model inference. RESYNC targets settings with background compute available but a strict per-observation response deadline.

## Silent Neuron Theory and Plasticity Preservation for Deep Reinforcement Learning in Adaptive Video Streaming

DevFeed: [Silent Neuron Theory and Plasticity Preservation for Deep Reinforcement Learning in Adaptive Video Streaming](<https://devfeed.tech/articles/silent-neuron-theory-and-plasticity-preservation-for-deep-reinforcement-learning-in-adaptive-video-streaming-131842.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2505.01584>)

Author: Zhiqiang He, Zhi Liu

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Neural Network](<https://devfeed.tech/topics/neural-network.md>), [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [deep](<https://devfeed.tech/tags/deep.md>), [deep-reinforcement-learning](<https://devfeed.tech/tags/deep-reinforcement-learning.md>), [learning](<https://devfeed.tech/tags/learning.md>), [neuron](<https://devfeed.tech/tags/neuron.md>), [plasticity](<https://devfeed.tech/tags/plasticity.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>), [streaming](<https://devfeed.tech/tags/streaming.md>), [theory](<https://devfeed.tech/tags/theory.md>), [video](<https://devfeed.tech/tags/video.md>), [video-streaming](<https://devfeed.tech/tags/video-streaming.md>)

### AI overview

The authors attribute difficulty adapting adaptive video streaming policies to heterogeneous bandwidth to neural network plasticity loss, which they say existing dormant-neuron metrics do not adequately capture. They propose Silent Neuron theory and Reset Silent Neuron (ReSiN), which resets neurons using both forward- and backward-propagation states to preserve plasticity; they also report a tighter performance bound under non-stationary conditions. In their implementation, ReSiN improved QoE by 178% when bandwidth alternated between high and low regimes, and by 1.6% on stationary real-world traces. These are reported results from the abstract; it provides no experimental setup or comparison details.

### Source excerpt

arXiv:2505.01584v5 Announce Type: replace-cross Abstract: Adaptive video streaming optimizes Quality of Experience (QoE) metrics by selecting appropriate bitrates according to varying network bandwidth and user demands. In practice, however, real-world network bandwidth often exhibits heterogeneity relative to training environments. Current methods predominantly tackle this problem through learning-based approaches designed to improve generalization performance. Our systematic investigation reveals a critical limitation: neural networks suffer from plasticity loss, significantly impeding their ability to adapt to heterogeneous network conditions. Through theoretical analysis of neural propagation mechanisms, we demonstrate that existing dormant neuron metrics inadequately characterize neural plasticity loss. To address this limitation, we have developed the Silent Neuron theory, which provides a more comprehensive framework for understanding plasticity degradation. Based on these theoretical insights, we propose the Reset Silent Neuron (ReSiN), which preserves neural plasticity through strategic neuron resets guided by both forward and backward propagation states. Moreover, we establish a tighter performance bound for ReSiN under non-stationary network conditions. In our implementation of an adaptive video streaming system, ReSiN shows substantial gains under non-stationary conditions in which the available bandwidth alternates between high and low regimes, improving the QoE by 178%, and it remains beneficial under stationary real-world traces, where it improves the QoE by 1.6%. Code will be available: https://tinyzqh.github.io/ReSiN/

## costream: A Streaming Evaluation Protocol for Wearable Fall Detection

DevFeed: [costream: A Streaming Evaluation Protocol for Wearable Fall Detection](<https://devfeed.tech/articles/costream-a-streaming-evaluation-protocol-for-wearable-fall-detection-132729.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2509.11789>)

Author: Timilehin B. Aderinola, Ilaria D'Ascanio, Luca Palmerini, Lorenzo Chiari, Jochen Klenk, Clemens Becker, Brian Caulfield, Georgiana Ifrim

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Streaming](<https://devfeed.tech/topics/streaming.md>)

Tags: [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [detection](<https://devfeed.tech/tags/detection.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [fall](<https://devfeed.tech/tags/fall.md>), [fall-detection](<https://devfeed.tech/tags/fall-detection.md>), [protocol](<https://devfeed.tech/tags/protocol.md>), [streaming](<https://devfeed.tech/tags/streaming.md>), [wearable](<https://devfeed.tech/tags/wearable.md>)

### AI overview

The evaluation protocol can change which wearable fall detector appears strongest: streaming and presegmented evaluation chose different methods in 6 of 20 model-window configurations, though they agreed on the overall best method. In streaming evaluation, a detector processes an unsegmented recording, aggregates window scores into a temporal confidence map, and emits timestamped events for comparison with annotated falls. The reported differences were associated with how false alarms clustered over time. Streaming evaluation also measures false alarm rate per monitoring time, detection delay, and threshold effects--deployment-relevant quantities that per-window evaluation does not capture. The comparison held the dataset, participant-wise splits, window construction, and fitted models fixed.

### Source excerpt

arXiv:2509.11789v2 Announce Type: replace Abstract: Wearable fall detectors are commonly evaluated on presegmented windows, where a fall window is extracted at an annotated impact and non-fall windows are sampled from fall-free signal. However, a deployed detector processes a continuous sensor stream in which the location of a fall is unknown. We specify a streaming evaluation protocol in which a detector traverses an unsegmented recording, aggregates window-level scores into a temporal confidence map, and emits timestamped events that are evaluated against the annotated fall. Holding the dataset, participant-wise splits, window construction and fitted models fixed, we compare presegmented and streaming evaluation on clinically verified real-world falls. The two protocols identify the same best-performing method overall, but select different methods in 6 of 20 model-window configurations, with the divergence associated with differences in the temporal clustering of false alarms. Streaming evaluation also exposes deployment-relevant quantities, including false alarm rate over elapsed monitoring time, detection delay, and the effect of the decision threshold, that are not captured by per-window evaluation. These results show that evaluation protocol can materially affect conclusions about wearable fall detectors and should be considered when assessing their suitability for continuous monitoring.

## MemInStream: Memorizing or Merely Recognizing? Rethinking Streaming Video Memory Evaluation

DevFeed: [MemInStream: Memorizing or Merely Recognizing? Rethinking Streaming Video Memory Evaluation](<https://devfeed.tech/articles/meminstream-memorizing-or-merely-recognizing-rethinking-streaming-video-memory-evaluation-132156.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.15640>)

Author: Zhenqi He, Dunyuan Xu, Yang Bo, Kien T. Pham, Yanghao Wang, Shuhan Li, Xintong Guo, Yuyang He, Fang Liu, Huankang Guan, Jingyan Chen, Rui Liu, Long Chen

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Streaming](<https://devfeed.tech/topics/streaming.md>), [benchmark overfitting machine learning](<https://devfeed.tech/topics/benchmark-overfitting-machine-learning.md>)

Tags: [benchmark](<https://devfeed.tech/tags/benchmark.md>), [capabilities](<https://devfeed.tech/tags/capabilities.md>), [computer-vision](<https://devfeed.tech/tags/computer-vision.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [integration](<https://devfeed.tech/tags/integration.md>), [memory](<https://devfeed.tech/tags/memory.md>), [protocol](<https://devfeed.tech/tags/protocol.md>), [retention](<https://devfeed.tech/tags/retention.md>), [retrieval](<https://devfeed.tech/tags/retrieval.md>), [shortcuts](<https://devfeed.tech/tags/shortcuts.md>), [state](<https://devfeed.tech/tags/state.md>), [streaming](<https://devfeed.tech/tags/streaming.md>), [streaming-video](<https://devfeed.tech/tags/streaming-video.md>), [tracking](<https://devfeed.tech/tags/tracking.md>), [updating](<https://devfeed.tech/tags/updating.md>), [validation](<https://devfeed.tech/tags/validation.md>), [video](<https://devfeed.tech/tags/video.md>)

### AI overview

MemInStream proposes evaluating streaming video memory separately from general question-answering accuracy, which can confuse forgetting with visual-understanding failures or allow correct answers from linguistic priors and recent observations. Its protocol links questions to temporally localized evidence and retains those solvable from that evidence but not from recent or post-evidence observations or text alone. The benchmark covers Episodic Retrieval, State Tracking & Updating, Multi-Evidence Integration, and Contextual Memory. The supplied abstract describes the design but gives no benchmark results.

### Source excerpt

arXiv:2611.15640v1 Announce Type: cross Abstract: Streaming video understanding aims to continuously process incoming visual observations, making the retention and retrieval of relevant information increasingly important as the video history grows. Although existing benchmarks include memory-related questions, question-answering accuracy alone does not reliably measure memory capabilities. Incorrect predictions may arise from failures in visual understanding rather than forgetting, whereas correct predictions may rely on linguistic priors or recent observations without requiring historical memory. To address these ambiguities, we introduce MemInStream, a benchmark dedicated to evaluating memory capabilities in streaming video understanding. We develop an evidence-controlled validation protocol that associates each question with temporally localized evidence intervals. The protocol validates evidence solvability and memory necessity by retaining questions that models can answer from the annotated evidence but cannot resolve using recent observations, post-evidence observations, or textual input alone. This helps disentangle memory limitations from understanding failures while reducing memory-independent shortcuts. MemInStream provides a systematic evaluation of memory capabilities through four major categories, covering Episodic Retrieval, State Tracking \& Updating, Multi-Evidence Integration, and Contextual Memory across diverse video scenarios. Overall, \textsc{MemInStream} provides a comprehensive and reliable benchmark for systematically evaluating memory capabilities in streaming video understanding.

## Live Streaming Sports to 32M Concurrent Viewers: Engineering Lessons Learned

DevFeed: [Live Streaming Sports to 32M Concurrent Viewers: Engineering Lessons Learned](<https://devfeed.tech/articles/live-streaming-sports-to-32m-concurrent-viewers-engineering-lessons-learned-128976.md>)

Original publisher: [Read original article](<https://www.scylladb.com/2026/11/11/live-streaming-ipl-jiohotstar/>)

Author: Cynthia Dunlop

Published: 2026-11-11T12:53:56Z

Content type: article

Language: en

Sources: [ScyllaDB](<https://devfeed.tech/sources/scylladb.md>)

Topics: [Production Engineering](<https://devfeed.tech/topics/production-engineering.md>), [Software Testing](<https://devfeed.tech/topics/software-testing.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>)

Tags: [2018](<https://devfeed.tech/tags/2018.md>), [2023](<https://devfeed.tech/tags/2023.md>), [alignment](<https://devfeed.tech/tags/alignment.md>), [analytics](<https://devfeed.tech/tags/analytics.md>), [api](<https://devfeed.tech/tags/api.md>), [app](<https://devfeed.tech/tags/app.md>), [application-servers](<https://devfeed.tech/tags/application-servers.md>), [architect](<https://devfeed.tech/tags/architect.md>), [audit](<https://devfeed.tech/tags/audit.md>), [community](<https://devfeed.tech/tags/community.md>), [concurrent](<https://devfeed.tech/tags/concurrent.md>), [conf](<https://devfeed.tech/tags/conf.md>), [cricket](<https://devfeed.tech/tags/cricket.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [indian](<https://devfeed.tech/tags/indian.md>), [live](<https://devfeed.tech/tags/live.md>), [live-streaming](<https://devfeed.tech/tags/live-streaming.md>), [platform](<https://devfeed.tech/tags/platform.md>), [sports](<https://devfeed.tech/tags/sports.md>), [streaming](<https://devfeed.tech/tags/streaming.md>), [uncategorized](<https://devfeed.tech/tags/uncategorized.md>), [video](<https://devfeed.tech/tags/video.md>)

### AI overview

JioHotstar prepared for the 2023 IPL final's 32 million concurrent viewers by planning for 50-75 million, protecting the video delivery path, and scaling authentication ahead of login surges that can peak on the tournament's first day. The team used a 200-point audit covering dependencies, redundancy, and timeouts; load-tested partners and redirected traffic when latency breached SLAs; and tested individual services, integrated systems, and full user journeys. Their latency budgets were about 10-15 ms for simple in-house calls, 40-50 ms for more complex work, and 5-10 ms for unavoidable third-party calls on the playback path, where fast failure can be preferable to waiting. The account recommends aligning event priorities, stress-testing systems, and removing nonessential synchronous dependencies from the critical path.

### Source excerpt

Ashutosh Agrawal shares how his team handled a record-setting level of concurrent viewers during the IPL cricket finals Live sports involve myriad unknowns, but one thing is certain: the live video feed must play flawlessly - and simultaneously - for all the fans tuning in. If some hear their neighbors cheering or groaning 5 seconds before each score, they're not going to have kind words for their streaming platform. Back in 2023, Ashutosh Agrawal and his JioHotstar (then JioCinema) engineering teammates were preparing for the Indian Premier League (IPL) cricket finals just as intensely as the athletes. While the players [...]

## SeqFold: Timestamp-Addressable Compressed States for User Action Sequence Backbones

DevFeed: [SeqFold: Timestamp-Addressable Compressed States for User Action Sequence Backbones](<https://devfeed.tech/articles/seqfold-timestamp-addressable-compressed-states-for-user-action-sequence-backbones-128346.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.13704>)

Author: Mianwei Zhou, Duo Zhang, Henry Tsai, Yuqi Zheng, Jian Li, Juexin Lin, Sujay Khandagale, Vijay Rajaram, Dong Jiao, Tao Cheng

Published: 2026-11-11T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Representation learning](<https://devfeed.tech/topics/representation-learning.md>), [Model Development](<https://devfeed.tech/topics/model-development.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>)

Tags: [actions](<https://devfeed.tech/tags/actions.md>), [channel](<https://devfeed.tech/tags/channel.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-si](<https://devfeed.tech/tags/cs-si.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [export](<https://devfeed.tech/tags/export.md>), [fusion](<https://devfeed.tech/tags/fusion.md>), [history](<https://devfeed.tech/tags/history.md>), [interface](<https://devfeed.tech/tags/interface.md>), [learning](<https://devfeed.tech/tags/learning.md>), [lookup](<https://devfeed.tech/tags/lookup.md>), [model](<https://devfeed.tech/tags/model.md>), [models](<https://devfeed.tech/tags/models.md>), [performance](<https://devfeed.tech/tags/performance.md>), [prediction](<https://devfeed.tech/tags/prediction.md>), [quantization](<https://devfeed.tech/tags/quantization.md>), [ranking](<https://devfeed.tech/tags/ranking.md>), [scale](<https://devfeed.tech/tags/scale.md>), [sequence](<https://devfeed.tech/tags/sequence.md>), [serving](<https://devfeed.tech/tags/serving.md>), [spaces](<https://devfeed.tech/tags/spaces.md>), [user](<https://devfeed.tech/tags/user.md>)

### AI overview

SeqFold exports states from a frozen long-history user-action backbone as compact codes addressable at target timestamps, so downstream systems can retrieve the state summarizing actions before each target without storing the full hidden-state timeline. It combines learned multi-channel fusion, channel-specific temporal decay, and Lookup-Free Quantization; task-specific decoders map the shared codes into embeddings for downstream models. A fused streaming implementation updates the fusion recurrence after each action but materializes states only at target timestamps. On large-scale proprietary datasets, the authors report 3,400x less storage per state while preserving most downstream ranking performance relative to the uncompressed model.

### Source excerpt

arXiv:2611.13704v1 Announce Type: new Abstract: Existing sequential user modeling methods mainly focus on improving the backbone encoder or learning task-specific user representations for a single prediction objective. We instead study how to export the multi-layer user states of a frozen long-history backbone as compact, timestamp-addressable representations for downstream training backfill and serving. In this setting, the user state at a given timestamp summarizes the user's actions before that time, so downstream systems need only the state aligned to each target timestamp rather than the full hidden-state timeline. To this end, we present \texttt{SeqFold}, a compressed-state interface that combines learned multi-channel fusion, channel-specific temporal decay, and Lookup-Free Quantization (LFQ) to produce compact discrete codes from frozen backbone states. Task-specific decoders then map these shared codes into flexible embedding spaces for downstream models. To make timestamp-addressable export practical at scale, we further develop a fused streaming implementation that updates the fusion recurrence after every action while materializing states only at target timestamps. On large-scale proprietary datasets, SeqFold reduces per-state representation storage by $3{,}400\times$ while preserving most of the downstream ranking performance of the uncompressed model, showing that SeqFold is an effective interface for reusing long-history sequence backbones in downstream models.

## Few-Step Streaming Video Generation via Condition Shifting

DevFeed: [Few-Step Streaming Video Generation via Condition Shifting](<https://devfeed.tech/articles/few-step-streaming-video-generation-via-condition-shifting-128389.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.14033>)

Author: Deyuan Liu, Jungang Li, Kaiwen Zheng, Kai Jiang, Jintao Chen, Yaole Wang, Ming Lin, Tao Lin, Jianfei Chen, Jun Zhu, Jintao Zhang

Published: 2026-11-11T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Generative AI](<https://devfeed.tech/topics/generative-ai.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>)

Tags: [apex](<https://devfeed.tech/tags/apex.md>), [condition](<https://devfeed.tech/tags/condition.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [distillation](<https://devfeed.tech/tags/distillation.md>), [generation](<https://devfeed.tech/tags/generation.md>), [gpu-memory](<https://devfeed.tech/tags/gpu-memory.md>), [motion](<https://devfeed.tech/tags/motion.md>), [streaming](<https://devfeed.tech/tags/streaming.md>), [streaming-video](<https://devfeed.tech/tags/streaming-video.md>), [training](<https://devfeed.tech/tags/training.md>), [video](<https://devfeed.tech/tags/video.md>), [video-generation](<https://devfeed.tech/tags/video-generation.md>)

### AI overview

APEX adapts few-step sampling to streaming video by shifting the text condition: the original condition drives generation, while a shifted one estimates the generated distribution's velocity field using the same generator parameters. This removes DMD's separate fake-score network while retaining S2-Avatar's frozen reference teacher, Self-Replay Forcing, and perceptual regularization. In reported evaluations on 180-second videos, APEX scored higher than S2-Avatar on Progressive-track instruction fulfillment (3.856 vs. 3.831) and visual aesthetics across both tracks (0.641898 vs. 0.632606). Human raters favored APEX most for instruction following (+30.0% net preference), with smaller advantages for consistency and video quality (+4.3% each); motion quality favored S2-Avatar, synchronization was mostly tied, and usability was unchanged. In separate matched-training measurements, APEX used 4.4% less peak allocated GPU memory per GPU and had 9.2% shorter mean iteration time than DMD.

### Source excerpt

arXiv:2611.14033v1 Announce Type: cross Abstract: Few-step sampling reduces the cost of streaming video generation. Distribution Matching Distillation (DMD) supervises few-step generators with a separate fake-score network, which adds training memory and computation and must track the evolving generator. We extend APEX to streaming video by using condition shifting to estimate the velocity field of the generated distribution within the generator. The original text condition selects generation, while a shifted condition selects fake estimation using the same parameters. This removes the independent fake network while retaining S2-Avatar's frozen reference teacher, Self-Replay Forcing, and perceptual regularization. We evaluate APEX on 180-second videos using three backbone evaluations per block and a one-step spatial Refiner, with a 70-case human study and a 320-case StreamAV-Bench comparison. Compared with S2-Avatar, APEX achieves higher video instruction fulfillment on the Progressive track (3.856 versus 3.831) and higher visual aesthetics across both tracks (0.641898 versus 0.632606). Human raters favor APEX most strongly on instruction following, with +30.0% net preference, followed by consistency and video quality at +4.3% each. Motion quality favors S2-Avatar, audiovisual synchronization is predominantly tied, and usability is unchanged. Separate measurements under matched training settings show 4.4% lower peak allocated GPU memory per GPU and 9.2% shorter mean iteration time than DMD.

## GigaEdge Spench: An SDK for Practical Streaming Speech Enhancement

DevFeed: [GigaEdge Spench: An SDK for Practical Streaming Speech Enhancement](<https://devfeed.tech/articles/gigaedge-spench-an-sdk-for-practical-streaming-speech-enhancement-128679.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.13243>)

Author: Artem Sokolov, Artem Khaidarov, Yana Lysova, Alexander Gladkikh, Vitaliy Yulis

Published: 2026-11-11T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [SDK](<https://devfeed.tech/topics/sdk.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>), [onnx runtime](<https://devfeed.tech/topics/onnx-runtime.md>)

Tags: [android](<https://devfeed.tech/tags/android.md>), [browsers](<https://devfeed.tech/tags/browsers.md>), [c-plus-plus](<https://devfeed.tech/tags/c-plus-plus.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [cross-platform](<https://devfeed.tech/tags/cross-platform.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [denoising](<https://devfeed.tech/tags/denoising.md>), [eess-as](<https://devfeed.tech/tags/eess-as.md>), [ios](<https://devfeed.tech/tags/ios.md>), [mobile-devices](<https://devfeed.tech/tags/mobile-devices.md>), [neural](<https://devfeed.tech/tags/neural.md>), [onnx](<https://devfeed.tech/tags/onnx.md>), [processing](<https://devfeed.tech/tags/processing.md>), [real-time](<https://devfeed.tech/tags/real-time.md>), [runtime](<https://devfeed.tech/tags/runtime.md>), [sdk](<https://devfeed.tech/tags/sdk.md>), [speech-enhancement](<https://devfeed.tech/tags/speech-enhancement.md>), [streaming](<https://devfeed.tech/tags/streaming.md>), [streams](<https://devfeed.tech/tags/streams.md>)

### AI overview

GigaEdge Spench is an open-source SDK for streaming speech enhancement across CPU servers, mobile devices, and browsers. Its configurable channels combine denoising, normalization and dynamics, sample-rate conversion, and acoustic echo cancellation where supported; normalization can also run without a neural model. It provides an 8 kHz INT8 denoiser and 48 kHz FP32 discriminative and hybrid models, with a shared C++ core and ONNX Runtime or TensorFlow Lite support. Independent streams can share model resources, and consecutive frames can be grouped into one inference call. Evaluation on EARS-WHAM v2 compares quality and processing costs with RNNoise and DPDFNet2. Tested native and Web configurations process files faster than real time, while larger chunks add input-collection delay. Reported SDK timings use ONNX Runtime.

### Source excerpt

arXiv:2611.13243v1 Announce Type: new Abstract: We present GigaEdge Spench, an open-source SDK for streaming speech enhancement on CPU servers, mobile devices, and in browsers. Applications can build enhancement channels from neural denoising, normalization and dynamics blocks, sample-rate conversion, and acoustic echo cancellation where supported. These stages can be combined, while normalization can also run without a neural model. The release includes an 8 kHz INT8 denoiser and discriminative and hybrid 48 kHz FP32 models. A shared C++ core supports ONNX Runtime and TensorFlow Lite through native interfaces, Android and iOS SDKs, and WebAssembly integration. Independent stream contexts share model resources, and consecutive frames from one stream can be processed in a single inference call. We evaluate the released models on EARS-WHAM v2 and compare their quality and processing costs with RNNoise and DPDFNet2. All tested native and Web configurations process files faster than real time, although larger chunks require more time to collect input. We also measure how many streams a shared model can process on a CPU. Our contribution is a configurable deployment system and an experimental study of how model choice, chunk size, and execution environment affect its use. All reported SDK timing uses ONNX Runtime. Index Terms: speech enhancement, streaming inference, cross-platform deployment, temporal frame grouping, computational efficiency

## Ninada: Sub-Clause Split-Graph VITS Streaming and Capability-Routed 24 kHz Speech Synthesis Across Twelve Scripts

DevFeed: [Ninada: Sub-Clause Split-Graph VITS Streaming and Capability-Routed 24 kHz Speech Synthesis Across Twelve Scripts](<https://devfeed.tech/articles/ninada-sub-clause-split-graph-vits-streaming-and-capability-routed-24-khz-speech-synthesis-across-twelve-scripts-127221.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.10035>)

Author: Hemanth HM

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Software Engineering](<https://devfeed.tech/sources/arxiv-software-engineering.md>)

Topics: [Speech Processing](<https://devfeed.tech/topics/speech-processing.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [audio](<https://devfeed.tech/tags/audio.md>), [bottlenecks](<https://devfeed.tech/tags/bottlenecks.md>), [browser](<https://devfeed.tech/tags/browser.md>), [capability](<https://devfeed.tech/tags/capability.md>), [checkpoint](<https://devfeed.tech/tags/checkpoint.md>), [client](<https://devfeed.tech/tags/client.md>), [convolution](<https://devfeed.tech/tags/convolution.md>), [covering](<https://devfeed.tech/tags/covering.md>), [cpus](<https://devfeed.tech/tags/cpus.md>), [cs-sd](<https://devfeed.tech/tags/cs-sd.md>), [cs-se](<https://devfeed.tech/tags/cs-se.md>), [dependency](<https://devfeed.tech/tags/dependency.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [desktop](<https://devfeed.tech/tags/desktop.md>), [graph](<https://devfeed.tech/tags/graph.md>), [scripts](<https://devfeed.tech/tags/scripts.md>), [speech](<https://devfeed.tech/tags/speech.md>), [speech-synthesis](<https://devfeed.tech/tags/speech-synthesis.md>), [split](<https://devfeed.tech/tags/split.md>), [streaming](<https://devfeed.tech/tags/streaming.md>), [synthesis](<https://devfeed.tech/tags/synthesis.md>)

### AI overview

Ninada reduces VITS speech synthesis startup latency by splitting an ONNX model in memory into prior, flow, and windowed vocoder sessions that stream with 12-frame halos and raised-cosine crossfades. Its 24 kHz system covers English, Classical Sanskrit, and ten Indian languages. Capability routing uses HiFi-GAN specialists for desktop/WebGPU and smaller distilled iSTFTNet specialists for mobile WASM, with a zero-dependency Klatt fallback. On 480 held-out studio utterances, it reports 3.22 dB 0-8 kHz MCD overall (3.09 dB for mobile iSTFTNet) and 6.07 dB error at 8-12 kHz. Reported audible startup is 89-155 ms on desktop WebGPU; on a physical Google Pixel 10 Pro XL, streaming is 538-649 ms with no underruns at 0.75-0.76x real-time factor.

### Source excerpt

arXiv:2611.10035v1 Announce Type: new Abstract: End-to-end VITS text-to-speech models synthesize natural speech in a single forward pass, yet standard ONNX exports execute monolithically over an entire clause, delaying Time-To-First-Audio (TTFA) to several seconds and spiking activation memory. Client-side deployment across Indic scripts faces four further bottlenecks: 138.8 MB float32 checkpoints per language strain browser caches, mobile WebGPU stalls for 6-14 s compiling convolution shaders while still running slower than real time ($1.44\times$ RTF), small WebAssembly pthread pools stall at barriers on big.LITTLE mobile CPUs, and mobile Web Audio pipelines drop frames or mute output under AudioWorklet contention and iOS WKWebView rules. We present Ninada, an open-source 24 kHz TTS system covering English, Classical Sanskrit, and ten Indian languages (sa, en, hi, kn, mr, bn, ta, te, ml, gu, pa, or) across a capability-routed 48-checkpoint fleet: 24 studio HiFi-GAN v1 specialists (27.9 MB W8A32+FP16 ONNX) for desktop/WebGPU and 24 distilled c8c8_384 iSTFTNet specialists (21.7 MB ONNX, $2.64\times$ fewer vocoder MACs via an exact NOLA dual-window ConvTranspose1d inverse STFT with zero complex ops) for mobile WASM, plus a 0.0 MB zero-dependency Klatt fallback. At runtime, Ninada splits a single ONNX file in memory in $< 15$ ms into prior, flow, and windowed vocoder sub-sessions streaming 12-frame halos with raised-cosine crossfades. On $N = 480$ held-out 24 kHz studio utterances, Ninada achieves 3.22 dB 0-8 kHz MCD (3.09 dB mobile iSTFTNet; 1.33 dB 24-voice distillation MCD, vs. 5.17 dB for the 1,526.8 MB 16 kHz Meta MMS fleet) and 6.07 dB 8-12 kHz error, delivering 89-155 ms audible startup on desktop WebGPU and 538-649 ms zero-underrun streaming at 0.75-0.76x RTF on a physical Google Pixel 10 Pro XL.

[Next page](<https://devfeed.tech/topics/streaming.md?cursor=WyIyMDI2LTExLTEwVDA1OjAwOjAwKzAwOjAwIiwgImEwZTViZjVhLTY0ZjItNDBlNC1iZTIyLTVhNzI0ZTMzNWUzYyJd>)