# lag Published articles for lag. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Adaptive Scheduling Pipeline for Multi-Instance Asynchronous Reinforcement Learning DevFeed: [Adaptive Scheduling Pipeline for Multi-Instance Asynchronous Reinforcement Learning]() Original publisher: [Read original article]() Author: Salah Chikhi, Luis Zuin, Entong Li, Zeng Li Published: 2026-11-09T05:00:00Z Content type: article Language: en Sources: [arXiv: Distributed Computing](), [arXiv: Machine Learning]() Topics: [llm-reasoning](), [Reinforcement learning](), [scheduling](), [LLM optimization](), [async](), [Large Language Model]() Tags: [adaptive](), [architecture](), [asynchronous](), [asynchronous-architecture](), [convergence](), [cs-ai](), [cs-dc](), [cs-lg](), [dispatch](), [domain](), [generation](), [inference](), [iteration](), [knowledge](), [lag](), [language](), [language-models](), [large-language-models-llms](), [learning](), [llms](), [pipeline](), [reinforcement-learning](), [scheduling]() ### AI overview ASP is an offline preprocessing scheduler designed to improve rollout utilization in multi-instance asynchronous LLM reinforcement learning. In one pass, it addresses padding bubbles caused by variable lengths within minibatches, skewness bubbles from long-tail outputs, and sorting bubbles that disrupt training-queue occupancy. The scheduler is integrated into an architecture that separates rollout from training while making policy lag observable and bounded. The abstract reports up to 3.22x end-to-end throughput improvement and up to 76.8% lower iteration time versus MindSpeed-RL, and up to 62.5% lower iteration time versus verl; scheduling alone reduces mean iteration time by 10.9%. It reports more robust scaling with added generation resources and no visible convergence degradation in the measured bounded-lag regime. An internal telecom-domain workflow reportedly cut training time from over three months to about two months. These are abstract-level claims; no experimental setup or detailed measurements are supplied here. ### Source excerpt arXiv:2611.08864v1 Announce Type: cross Abstract: We present the Adaptive Scheduling Pipeline (ASP), an offline rollout scheduler for multi-instance asynchronous reinforcement learning (RL) of large language models (LLMs). ASP jointly reduces three sources of rollout underutilization - padding bubbles from intra-minibatch length heterogeneity, skewness bubbles from long-tail outputs, and sorting bubbles from fluctuations in training-queue occupancy induced by length-sorted dispatch - through a single offline preprocessing pass. This enables efficient multi-instance rollout that overcomes the early scaling plateau of single-instance inference while smoothing the arrival of generated samples at the trainer. We integrate the scheduler into an asynchronous architecture that decouples rollout from training while keeping policy lag observable and bounded. Complete ASP improves end-to-end throughput by up to 3.22 times and reduces iteration time by up to 76.8% relative to MindSpeed-RL, while reducing iteration time by up to 62.5% relative to verl. Within ASP, scheduling reduces mean iteration time by 10.9%. ASP also scales more robustly with additional generation resources than these baselines and shows no visible convergence degradation in the measured bounded-lag regime. ASP has since been used in an internal telecom-domain RL post-training workflow, reducing the reported training period from more than three months to about two months and producing a domain-enhanced model for knowledge answering and tool calling. ## TimeOmni-Live: Continuous & Live Time Series Understanding with Long-Term Memory DevFeed: [TimeOmni-Live: Continuous & Live Time Series Understanding with Long-Term Memory]() Original publisher: [Read original article]() Author: Tong Guan, Alexander Chemeris, Yiqing Guo, Warren Jin, Zhao Li, Shirui Pan, Ming Jin Published: 2026-11-09T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Time Series](), [Temporal data](), [streaming data analytics]() Tags: [continuous](), [cs-lg](), [edge](), [efficiency](), [evaluations](), [hybrid](), [inference](), [jetson](), [lag](), [language-models](), [latency](), [live](), [llms](), [memory](), [model](), [models](), [offline](), [open](), [series](), [time](), [time-series]() ### AI overview TimeOmni-Live targets continuous time-series streams with persistent historical context, addressing the limits of offline models that process fixed windows independently. Its design combines native temporal encoding, hybrid memory for exact evidence and long-term task states, and elastic synchronization intended to reduce inference lag. The accompanying LiveTS-Suite contains over 5,000 streams and 2.2 billion timesteps, with 15,000 response events across 18 tasks and six capabilities; evaluation jointly measures timeliness, accuracy, and computational efficiency. The abstract reports a score of 90.0 versus 79.3 for Opus 5 with a 4B-parameter model, p95 latency 206 times lower than its backbone, and real-time local operation demonstrated on a Jetson edge device. ### Source excerpt arXiv:2611.09502v1 Announce Type: new Abstract: Existing time series language models predominantly follow an offline, one-shot paradigm: they process fixed-length inputs independently and do not retain past states. This static approach fundamentally misaligns with the continuous nature of real-world streaming applications and fails to maintain the historical context necessary for cross-window reasoning. To bridge this gap, we introduce Continuous & Live Time Series Understanding (CL-TSU), a novel paradigm that shifts the focus toward maintaining a persistent, stateful understanding over open-ended data streams. To systematically evaluate this paradigm, we present LiveTS-Suite, the first comprehensive data suite tailored for CL-TSU, supporting data generation, model training, and systematic evaluation across 18 tasks spanning six core capabilities. It comprises over 5K streams totaling 2.2B timesteps, annotated with 15K specific response events. We further design an evaluation protocol that jointly assesses response timeliness, content accuracy, and computational efficiency. Building on this foundation, we propose TimeOmni-Live, the first live time series understanding model. It integrates a hybrid time series encoding for natively temporal input, a stream-persistent hybrid memory mechanism to effectively preserve exact historical evidence and long-term task states, and an elastic stream-inference synchronization strategy to reduce processing lag. Evaluations show that TimeOmni-Live establishes a strong baseline for CL-TSU, outperforming frontier LLMs (90.0 vs. 79.3 for Opus 5) with only 4B parameters and achieving 206$\times$ lower p95 latency than its backbone. Deployment on a Jetson edge device further shows its ability to support real-time continuous temporal understanding locally at the edge. ## Scratchpad Patching: Decoupling Compute from Patch Size in Byte-Level Language Models DevFeed: [Scratchpad Patching: Decoupling Compute from Patch Size in Byte-Level Language Models]() Original publisher: [Read original article]() Author: Lin Zheng, Vasilisa Bashlovkina, Timothy Dozat, Dan Garrette, Laura Rimell, Joshua Maynez Published: 2026-11-09T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Chinchilla scaling law](), [LLM evaluation / benchmarking]() Tags: [cache](), [code](), [cs-cl](), [cs-lg](), [efficiency](), [entropy](), [inference](), [lag](), [language-models](), [patching](), [quality](), [size]() ### AI overview Scratchpad Patching (SP) addresses patch lag in byte-level language models: predictions within a patch otherwise rely on a stale representation until that patch is fully observed, and larger patches widen this lag. SP inserts transient scratchpads to aggregate observed bytes and refresh patch-level context. It triggers them using next-byte prediction entropy, allocating compute selectively to information-dense regions and allowing inference-time compute to be adjusted after training. In experiments on natural language and code, SP improved quality at the same patch size. At 16 bytes per patch, SP models matched or closely approached the byte-level baseline on downstream evaluations, with a 16x smaller KV-cache over patches and roughly 3-4x less inference compute than the baseline. ### Source excerpt arXiv:2605.09630v2 Announce Type: replace-cross Abstract: Tokenizer-free language models eliminate the tokenizer step of the language modeling pipeline by operating directly on bytes; patch-based variants further aggregate contiguous byte spans into patches for efficiency. However, the average patch size chosen at the model design stage governs a tight trade-off: larger patches reduce compute and KV-cache footprint, but degrade modeling quality. We trace this trade-off to patch lag: until a patch is fully observed, byte predictions within it must rely on a stale representation from the previous patch to preserve causality; this lag widens as patches grow larger. We introduce Scratchpad Patching (SP), which inserts transient scratchpads inside each patch to aggregate the bytes seen so far and refresh patch-level context for subsequent predictions. SP triggers scratchpads using next-byte prediction entropy, selectively allocating compute to information-dense regions and enabling post-hoc adjustment of inference-time compute. Across experiments on natural language and code, SP improves model quality at the same patch size; for example, even at $16$ bytes per patch, SP-augmented models match or closely approach the byte-level baseline on downstream evaluations while using a $16\times$ smaller KV-cache over patches and requiring roughly $3$-$4\times$ less inference compute than the byte-level baseline. ## HeapDeck: an on-call monitor for PostgreSQL on iPhone and iPad DevFeed: [HeapDeck: an on-call monitor for PostgreSQL on iPhone and iPad]() Original publisher: [Read original article]() Published: 2026-11-07T00:00:00Z Content type: release Language: en Sources: [PostgreSQL news]() Topics: [PostgreSQL](), [database monitoring](), [Monitoring & Alerting](), [dashboards](), [iphone](), [ssh](), [ipad](), [TLS (Transport Layer Security)]() Tags: [alerting](), [alerts](), [app-store](), [blocking](), [canada](), [client](), [counts](), [free](), [ipad](), [iphone](), [lag](), [locks](), [monitor](), [native](), [on-call](), [pg-stat-statements](), [pinned](), [postgresql](), [ssh](), [tls]() ### AI overview HeapDeck is a native iPhone and iPad monitor for PostgreSQL 13-19, designed to help on-call engineers inspect database health while the app is open. It presents a health verdict and ten tiles; its Locks screen builds a blocking tree from pg_blocking_pids() and shows how many sessions stopping the root blocker would free. Other views cover activity, transaction ID wraparound, dead tuples, replication lag and slots, and query statistics. Connections are direct or use an SSH tunnel, with TLS and pinned fingerprints; no vendor server sits between the phone and database. HeapDeck sends no alerts, so it complements existing alerting. Monitoring one server is free; a one-time HeapDeck Pro purchase adds more servers and actions including cancel, terminate, VACUUM, dropping a slot, and writes. ### Source excerpt HeapDeck is a new native app for engineers on call for PostgreSQL 13 to 19, on the App Store for iPhone and iPad. It opens on one health verdict with ten tiles. The Locks screen draws the tree from pg_blocking_pids() with the root blocker on top and counts the sessions that stopping it would free. Other screens cover pg_stat_activity, transaction ID wraparound and dead tuples, replication lag and slots, and pg_stat_statements. The app uses a direct connection or an SSH tunnel, with TLS and pinned fingerprints. No vendor server sits between the phone and the database. It works while open and sends no alerts, so it goes next to your existing alerting. Monitoring one server is free. HeapDeck Pro is a one-time in-app purchase that adds more servers and the actions: cancel, terminate, VACUUM, dropping a slot and writes. https://heapdeck.app Postgres, PostgreSQL and the Slonik Logo are trademarks or registered trademarks of the PostgreSQL Community Association of Canada, and used with their permission. HeapDeck is an independent client and is not affiliated with, endorsed by, or sponsored by the PostgreSQL Community Association of Canada, PostgreSQL Global Development Group, PostgreSQL Project, PostgreSQL Core Team, or PostgreSQL Community. ## Ordering Multivariate Time Series by Volatility Memory: A Temporal Convolutional Approach to Volatility Heredity DevFeed: [Ordering Multivariate Time Series by Volatility Memory: A Temporal Convolutional Approach to Volatility Heredity]() Original publisher: [Read original article]() Author: Kaoutar El Azaar, Mohamed Hamlich, Moncef Garouani, Franck Ravat, Asmae Chakir, Mohamed Tabaa Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Time Series](), [World models]() Tags: [cryptocurrency](), [cs-lg](), [finance](), [graph](), [history](), [lag](), [learning](), [linear](), [markets](), [matrix](), [memory](), [ordering](), [persistence](), [quality](), [series](), [structured](), [synthetic](), [systems](), [temporal](), [time-series](), [variables]() ### AI overview QS-TCN orders multivariate time series using volatility memory to uncover directional relationships that conditional-mean analysis may miss. It combines persistence estimates from quasi-maximum likelihood and temporal modes from the Matrix Pencil method into structured temporal convolution filters, alongside learned filters, then aggregates pairwise directional evidence into a global precedence order. That order does not identify a causal graph. In four synthetic volatility regimes, QS-TCN had higher mean ordering quality than an external lead-lag method and its squared-input adaptation, though its advantage over the adaptation was uncertain in two regimes. It was also evaluated on Finance and fMRI benchmarks with known reference relations; currency, gold, and cryptocurrency applications are reported descriptively. ### Source excerpt arXiv:2611.07746v1 Announce Type: new Abstract: Volatility dynamics can carry information about temporal relationships that is not apparent in the conditional mean. Extracting this information is difficult because the persistence of each variable can obscure its dependence on other variables. This paper presents QS-TCN, a method for ordering multivariate time series that combines statistical estimates of volatility memory with temporal convolutional learning. Persistence estimates obtained by quasi-maximum likelihood and temporal modes extracted by the Matrix Pencil method define structured filters within the network. Together with learned filters, they represent the history of the target and support comparisons of directional evidence between variables. These comparisons are combined into a global order of precedence, which does not identify a causal graph. The method is evaluated on synthetic systems with linear and nonlinear volatility interactions and on the Finance and fMRI benchmarks, whose reference relations are known. Across the four synthetic volatility regimes studied, QS-TCN reaches a higher mean ordering quality than an external lead--lag ordering and its squared-input adaptation, although the gain over the adaptation is uncertain in two regimes. Applications to currency, gold and cryptocurrency markets are reported descriptively. ## Zero Mismatch, Simple Async RL: REINFORCE Zero, DeepSeek-V4.1 and Kimi K3.1 DevFeed: [Zero Mismatch, Simple Async RL: REINFORCE Zero, DeepSeek-V4.1 and Kimi K3.1]() Original publisher: [Read original article]() Author: Ao Shen (Inferact), Jian Hu (NVIDIA Research), Yan Bai (NVIDIA DevTech), Shekhar Pandey (AMD), and the REINFORCE Zero authors Published: 2026-11-06T00:00:00Z Content type: article Language: en Sources: [vLLM Blog]() Topics: [LLM optimization](), [async](), [llm-reasoning](), [quantization](), [grpo]() Tags: [2](), [3](), [7-4](), [8k](), [about](), [ai-engineering](), [align](), [amd](), [asynchronous](), [deepseek](), [deepseek-v4-1-flash](), [flash](), [kernels](), [kimi](), [lag](), [llm-inference](), [model-serving](), [open-source-ai](), [reinforcement-learning](), [rl](), [sampling](), [training](), [trains](), [v4-1-flash](), [vllm-blog](), [zero]() ### AI overview REINFORCE Zero argues that asynchronous RL can train stably on stale samples when rollout and learner kernels are aligned, leaving unclipped token level importance sampling to correct weight lag. The paper reports stable training through 20,000 updates and learning at policy lag 64; in one comparison, importance weight deviation peaked at 1.50 times its median with aligned kernels versus 442.54 times with standard kernels. Without alignment, training drifted and a sequence gate rejected 24.48% of sequences. Alignment has a speed cost: one OSWorld setup took 6.8% longer per step than standard kernel IS with a gate. The excerpt also reports zero-mismatch deployments for DeepSeek-V4.1-Flash and Kimi K3.1, including AMD MI355X runs, while noting ongoing performance work and that long asynchronous runs on the flagship models remain future work. ### Source excerpt Asynchronous RL trains on stale weights. Align rollout and training kernels bit for bit, and unclipped importance sampling corrects that lag. Plus zero-mismatch RL for DeepSeek-V4.1-Flash, Kimi K3.1 and AMD MI355X. ## MongoDB CDC connector for ClickPipes is now generally available DevFeed: [MongoDB CDC connector for ClickPipes is now generally available]() Original publisher: [Read original article]() Author: Marta Paes Published: 2026-11-04T17:46:41Z Content type: release Language: en Sources: [ClickHouse Blog]() Topics: [MongoDB](), [streaming data analytics](), [releases](), [Prometheus](), [OpenAPI Specification]() Tags: [cdc](), [lag](), [metrics](), [mongodb](), [openapi](), [parallel](), [resource](), [structured](), [terraform]() ### AI overview MongoDB CDC ClickPipes is generally available for ClickHouse Cloud, using MongoDB Change Streams for near-real-time replication. GA adds automatic parallel snapshots partitioned by `_id`, with sampled boundaries to balance uneven collections. Structured ingestion is optional: it infers a schema and maps top-level fields to typed columns; unknown fields and type mismatches are retained in `_peerdb_malformed_data`. Deeper flattening is not yet supported. Users can configure target fields, partitioning, sorting keys, enable collection pre-images for deletes, or filter deletes to retain history. New CPU, memory, network, and source, destination, and end-to-end lag metrics are available in the UI and Prometheus endpoint. OpenAPI and Terraform management require Terraform v3.14.0 or later. Billing began November 4, 2026; the connector is available in all ClickHouse Cloud service tiers. ### Source excerpt The MongoDB CDC connector for ClickPipes is now generally available, with parallel snapshotting, structured ingestion into typed columns, lag and resource metrics, and Terraform and OpenAPI support. ## Slow Job Alerts and Replica Commit IDs in Laravel 13.40 DevFeed: [Slow Job Alerts and Replica Commit IDs in Laravel 13.40]() Original publisher: [Read original article]() Author: Paul Redmond Published: 2026-11-03T23:53:01Z Content type: release Language: en Sources: [Laravel]() Topics: [Laravel](), [read-replica]() Tags: [alerts](), [commit](), [ids](), [job](), [jobs](), [lag](), [laravel](), [laravel-releases](), [middleware](), [news](), [queue](), [read-replica](), [replica](), [replica-lag](), [scheduled](), [scheduled-tasks](), [tasks]() ### AI overview Laravel 13.40.0 adds runtime alerts for queue jobs and scheduled tasks: plain numeric thresholds are milliseconds for jobs and seconds for scheduled tasks, and callbacks run after job attempts or task completion/failure. Database commit IDs let applications check whether replicas have received writes, with optional waiting; support is specified for MySQL with GTIDs, MariaDB with binary logging, and PostgreSQL, though PostgreSQL waiting requires version 19 and other drivers throw. Middleware can now be inserted around built-in middleware, and strict request input casting returns the default for invalid values; strict booleans accept only listed common true/false forms. The release also secures process-driver concurrency results with signed frames, reconnects after transaction connection loss, and rolls back nested transactions to savepoints on lock timeouts or deadlocks. ### Source excerpt Laravel 13.40 adds callbacks for slow queue jobs and scheduled tasks, commit IDs to check read replica lag, and insertBefore() and insertAfter() for middleware. The post Slow Job Alerts and Replica Commit IDs in Laravel 13.40 appeared first on Laravel News. Join the Laravel Newsletter to get Laravel articles like this directly in your inbox. ## A Hybrid Visitor Prediction Framework using Interpretable Temporal features, Integrating IQR and Anomaly Detection in Lakes in Nebraska DevFeed: [A Hybrid Visitor Prediction Framework using Interpretable Temporal features, Integrating IQR and Anomaly Detection in Lakes in Nebraska]() Original publisher: [Read original article]() Author: Israt Zarin, Christopher Chizinski Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Feature Engineering](), [Data science and engineering](), [Artificial Intelligence]() Tags: [anomaly-detection](), [cross-validation](), [cs-lg](), [features](), [hybrid](), [lag](), [metrics](), [model](), [models](), [outlier-detection](), [prediction](), [regression](), [temporal](), [tourism]() ### AI overview The framework predicts daily visitation at Nebraska lakes using spatial and temporal features, including interpretable temporal and lag features, and combines strong base models in hybrid ensembles. The reported results indicate that removing anomalies reduced MAE, MSE, RMSE, and MAPE, and that anomaly detection outperformed IQR while preserving data. In anomaly-based cross-validation, the voting ensemble improved prediction errors over the MLP (256,128,64): MAE by 9.82%, MSE by 3.12%, RMSE by 1.76%, and MAPE by 6.77%. Compared with the best IQR-based hybrid, the best anomaly-based hybrid had lower cross-validation errors by 9.81%, 34.30%, 18.99%, and 0.66% on those metrics. The abstract says lag features had some positive effects; it does not quantify them. ### Source excerpt arXiv:2611.03486v1 Announce Type: new Abstract: For effective resource management, infrastructure planning, and sustainable tourism, it is really important to predict accurate visitors. This paper estimates the daily visitation of Nebraska lakes using spatial, basic temporal, and interpretable temporal features. Moreover, we evaluate the base models with multiple metrics. Using the best three models, our method will ensemble hybrid models and choose the best hybrid technique. Further, we investigate the impact of lag features and how the models work with different outlier detection techniques such as Interquartile Range (IQR) and anomaly detection. The models are evaluated using regression metrics (MAE, MSE, RMSE, MAPE, and R2). Experiment results show that, after removing anomalies, MAE, MSE, RMSE, and MAPE are reduced. Adding lag features also shows some positive effects on the results. Moreover, anomaly detection techniques perform better than IQR in improving performance while preserving data. The reported anomaly-based cross-validation results favored the voting ensemble over the compared base models on the evaluated metrics. In anomaly-based cross-validation, VOTE improved prediction errors compared with MLP (256,128,64); improvements varied in the single-split comparison. Specifically, the hybrid model reduced MAE by 9.82%, MSE by 3.12%, RMSE by 1.76%, and MAPE by 6.77%. In the reported cross-validation comparison, the best anomaly-based hybrid model had lower MAE, MSE, RMSE, and MAPE than the best IQR-based hybrid model by 9.81%, 34.30%, 18.99%, and 0.66%, respectively. ## On the Origin and Evolution of Training-Inference Mismatch in LLM Reinforcement Learning DevFeed: [On the Origin and Evolution of Training-Inference Mismatch in LLM Reinforcement Learning]() Original publisher: [Read original article]() Author: Yu Li, Yufa Zhou, Tian Lan Published: 2026-11-02T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Machine Learning]() Topics: [Reinforcement learning](), [llm-reasoning](), [Large Language Model](), [Machine learning]() Tags: [asynchronous](), [bias](), [centering](), [cs-ai](), [cs-lg](), [dlc](), [evolution](), [feedback](), [inference](), [inference-engine](), [kl-divergence](), [lag](), [language-models](), [large-language-models](), [llm](), [numerics](), [policy](), [qwen3](), [reinforcement-learning](), [sampling](), [trainer](), [training]() ### AI overview The abstract separates training-inference mismatch into engine numerical discrepancy and trainer policy lag. Numerical discrepancy remains approximately constant and causes nearly fixed update bias, whose leading component is removed by centering scores at each prefix. Policy lag grows through a feedback loop in which bias from one update enlarges the next lag. Importance weighting interrupts this loop only while lag ratios remain concentrated; on Qwen3-1.7B it fails when their effective sample size drops below one half. The proposed discrepancy-lag correction (DLC) centers numerical discrepancy, clips lag ratios against sampler weights, and resynchronizes the sampler below that threshold. With a stale sampler, DLC kept divergence from on-policy training within twice the divergence between independent on-policy runs on Qwen3-0.6B and Qwen3-1.7B. Resynchronization also prevented collapses missed by per-token corrections and by KL measured on sampler prefixes. ### Source excerpt arXiv:2610.40945v1 Announce Type: new Abstract: On-policy reinforcement learning for large language models assumes that training responses are sampled from the current policy. In practice, an inference engine samples the responses and a separate trainer computes the gradients, and the \textbf{training-inference mismatch} between the two makes the training data off-policy. The mismatch has two sources, the numerics of the engine and the stale weights of asynchronous sampling, and a correction that stabilizes training under one source can fail under the other. \textit{We ask which property of a mismatch source decides the correction it needs.} Evaluating the trainer at the sampler's weights splits each sampled-token error exactly into a \textit{numerical discrepancy} of the engine and a \textit{policy lag} of the trainer, and we follow both through training. The numerical discrepancy stays approximately constant, so it biases the updates by a nearly fixed amount, and centering the score at each prefix removes its leading bias. The policy lag is produced by the trainer's own updates: the bias it leaves enters the next update, which enlarges the next lag and closes a feedback loop. Importance weighting cuts this loop only while the lag ratio stays concentrated, and on Qwen3-1.7B it fails at the step where the effective sample size of the ratio falls below one half. From this account we derive the \textbf{discrepancy-lag correction} (DLC), which centers the numerical discrepancy, clips the lag ratio against the sampler's weights and resynchronizes the sampler when this effective sample size falls below one half. Under a stale sampler, DLC keeps the KL divergence from on-policy training within twice the KL between two independent on-policy trainings on Qwen3-0.6B and Qwen3-1.7B, and its resynchronization prevents collapses that the per-token corrections leave and that a KL measured on the sampler's prefixes does not detect. ## Alerting in the Debezium Platform DevFeed: [Alerting in the Debezium Platform]() Original publisher: [Read original article]() Author: Indra Raj Shukla Published: 2026-11-02T00:00:00Z Content type: news Language: en Sources: [Debezium]() Topics: [Monitoring & Alerting](), [Prometheus](), [Apache Kafka]() Tags: [alerting](), [alerts](), [backend](), [badge](), [channel](), [collector](), [debezium](), [lag](), [monitoring](), [notifications](), [pipelines](), [platform](), [queue](), [snapshot]() ### AI overview Debezium Management Platform alert rules apply comparisons and thresholds to monitoring panels and evaluate globally, creating a separate incident for each pipeline that breaches a rule, including pipelines added later. Rules choose a panel, a reduction function (`last`, `avg`, `min`, `max`, or `sum`), a comparison and threshold, a hold duration, severity, and optional notification channels. The Platform evaluates enabled rules every 60 seconds by default; each firing and resolution sends one notification, with no repeats while an alert remains active. Alerts always appear in the UI. Optional reusable email and webhook channels support test notifications; webhooks send a versioned payload for both firing and resolved states. The Events page lists incidents, while a header badge counts active Critical and Warning alerts; Info alerts are excluded. Resolved incidents are removed after a default 30-day retention period, while firing incidents are retained. ### Source excerpt The Debezium Management Platform has been able to show you how your pipelines are doing for a while now. The monitoring dashboard tells you the event rate, the source lag, the queue utilization, whether the snapshot finished, and so on. But it is passive as it only helps while somebody is looking at it. In his mid-year Platform update, Mario listed alerting as one of the next things we wanted to tackle. That work has now landed. The Platform can now watch those same panels for you and tell you when one of them crosses a line you care about. This post walks through what alerting looks like in the Platform, how a rule is defined, where notifications go, how an alert moves from quiet to firing, and where to review it all afterwards. Read more... ## Planned Restlessness: Delayed Meta Bandits with Budgets DevFeed: [Planned Restlessness: Delayed Meta Bandits with Budgets]() Original publisher: [Read original article]() Author: Sudipto Guha Published: 2026-10-30T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Reinforcement learning](), [thompson sampling](), [Combinatorial optimization]() Tags: [approximation](), [bandits](), [bayesian](), [cs-ds](), [cs-lg](), [explore](), [future](), [information](), [lag](), [lower-bound](), [lp-relaxation](), [optimization](), [reward](), [risk](), [sampling](), [thompson-sampling]() ### AI overview The paper develops a planning framework for episodic Bayesian bandits where arm qualities are redrawn from unknown recurring priors, observations may be delayed, and plays face hard caps and budgets. Its LP relaxation values experiments by immediate reward, resource use, and the future decisions enabled by the resulting learning state; dual prices yield per-arm programs and policy certificates. The framework includes hierarchical Thompson sampling, risk-sensitive objectives, purchased observations, and buyout options, while execution accounts for preserving or losing planned learning transitions under constraints. With stationary priors, finite reward alphabets, bounded report lag, and nonbinding budgets, posterior sampling over priors has an O(√N) bound on the cost of prior uncertainty across N episodes (other parameters fixed). In the separate undelayed single-episode setting with each arm capped at T/X plays, the stated LP-relative approximation is 1+(2πX)^−1/2+O(X^−1), with a matching lower bound on the leading term. ### Source excerpt arXiv:2610.38955v1 Announce Type: new Abstract: We study episodic finite-horizon Bayesian bandits, or meta-bandits, with independent arms. Each episode draws fresh arm qualities from unknown recurring priors. Observations arrive after known delays, and plays obey hard per-episode caps and total budgets. A play earns reward and informs both the current quality and the recurring prior: the learner must explore, use, and generalize. We formulate a common planning optimization over learning trajectories. Its LP relaxation values experiments through current reward, resource consumption and continuation: the value of future decisions from the learning state produced. Dual prices separate the arms into local programs and certify every admissible policy, including hierarchical Thompson sampling. Execution enforces the hard constraints while preserving the promised learning transitions or accounting for their loss. The framework also represents risk-sensitive objectives and prices purchased observations and buyout options. Delayed reports create planned restlessness: an arm's information changes while it is not played, through reports of its own earlier plays arriving on a known calendar. For stationary priors, finite reward alphabets, bounded report lag and nonbinding budgets, posterior sampling over the recurring priors bounds the price of not knowing them by $O(\sqrt{N})$ over $N$ episodes, with other parameters fixed. For one undelayed $T$-step episode with arms capped at $T/X$ plays each, we obtain the LP-relative approximation factor $1+(2\pi X)^{-1/2}+O(X^{-1})$, improving previous factors, with a matching lower bound for the leading term. ## ATEMPO: Amortized Transfer Entropy Model Per-lag Over-Time DevFeed: [ATEMPO: Amortized Transfer Entropy Model Per-lag Over-Time]() Original publisher: [Read original article]() Author: Emma Andrews, Konstantinos Kontras, Keane Ong, Maarten De Vos, Gianmarco Mengaldo Published: 2026-10-28T04:00:00Z Content type: news Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Time Series](), [Machine learning](), [bci]() Tags: [2](), [3](), [about](), [abstract](), [all](), [arxiv](), [based](), [benchmark](), [best](), [brain](), [but](), [can](), [computer](), [cs-ai](), [cs-it](), [datasets](), [delay](), [entropy](), [lag](), [math-it](), [model](), [over](), [time]() ### AI overview ATEMPO estimates transfer entropy across source and target history depths and over time with one diffusion-based score network. It samples depths during training and conditions on temporal position to produce a full TE surface without refitting; differencing that surface identifies when information arrives. For a 6-by-3 depth grid, the abstract reports one fit versus 18 separate fits for existing estimators. On synthetic data, ATEMPO had the lowest mean relative error among baselines and achieved 0.98±0.04 lag-detection accuracy on datasets with known lags. It maintained perfect lag detection across evaluated window lengths on three time-varying-lag systems. Selected lags also improved reported forecasting results: 7.25% lower MSE on WEATHER-5K, and 120 inputs beat a 240-feature default by 4.6%; across three BCI tasks, mean R² improved 4.1% over the strongest competing baseline. ### Source excerpt arXiv:2610.36536v1 Announce Type: new Abstract: When one time series informs another, the delay at which that information arrives determines if models can utilize this relationship effectively. A source may be strongly predictive at one delay but uninformative at neighboring lags. Therefore, fixed input windows may obscure this structure by bundling informative and irrelevant history together. Transfer entropy (TE) can recover these delays by measuring how much a source's past tells us about a target beyond the target's own history. However, existing TE estimators require separate fits across source and conditioning depths, with further estimation on local windows to track delays over time. We introduce ATEMPO, a diffusion-based transfer entropy estimator amortized over lags and time. ATEMPO samples source and conditioning depths during training and conditions explicitly on temporal position, allowing a single score network to estimate the full TE surface without refitting. Differencing this surface recovers when information arrives. Across synthetic datasets with ground-truth TE, ATEMPO achieves the lowest mean relative error among baselines, and on datasets with known lags, ATEMPO achieves $0.98\pm0.04$ lag-detection accuracy. ATEMPO requires a single fit across source depths $K=6$ and target depths $L=3$, whereas existing estimators require 18 separate fits to evaluate the same $K \times L$ grid. On three systems with time-varying lag, ATEMPO is the only estimator to maintain perfect lag-detection accuracy across all evaluated window lengths. These recovered delays improve downstream forecasting: ATEMPO-selected inputs reduce MSE on WEATHER-5K by 7.25%, while 120 selected inputs outperform the benchmark's 240-feature default by 4.6%. Across three brain-computer interface tasks, using ATEMPO-selected lags achieves the highest or equal-best $R^2$, improving mean $R^2$ by 4.1% relative to the strongest competing baseline. ## BadTime: An Effective Backdoor Attack on Multivariate Long-Term Time Series Forecasting DevFeed: [BadTime: An Effective Backdoor Attack on Multivariate Long-Term Time Series Forecasting]() Original publisher: [Read original article]() Author: Kunlan Xiang, Haomiao Yang, Wenbo Jiang, Haoxin Wang, Shiyue Huang, Shaofeng Li, Yijing Liu, Ji Guo Published: 2026-10-28T04:00:00Z Content type: article Language: en Sources: [arXiv: Cryptography and Security]() Topics: [Adversarial attacks](), [Time Series](), [ai security](), [Neural Network]() Tags: [attack](), [backdoor](), [challenges](), [climate](), [cs-cr](), [data](), [data-driven](), [distributed](), [forecasting](), [future](), [horizon](), [i](), [improvement](), [influence](), [lag](), [learning](), [security](), [series](), [time](), [time-series](), [time-series-forecasting]() ### AI overview BadTime is a backdoor attack designed to steer multivariate long-term time-series forecasts toward a target pattern while keeping clean forecasting error nearly unchanged. It uses a distributed, lag-aware trigger based on inter-variable correlations and data-driven initialization to preserve its effect over long horizons. To make sparse backdoor signals learnable, it mixes label-similar and label-dissimilar poisoned samples and uses a decoupled objective that up-weights the signal; the abstract reports reliable learning at a 1% poisoning rate. Experiments report an attackable horizon of up to 720 timesteps, versus at most 12 for the compared state-of-the-art attack, 57% lower target-variable MAE on average, and under 1% change in average clean forecasting error. These are claims from the supplied abstract. ### Source excerpt arXiv:2508.04189v3 Announce Type: replace Abstract: Multivariate long-term time series forecasting (MLTSF) models are increasingly deployed in critical domains such as climate, finance, and transportation. Despite their growing importance, the security of MLTSF models against backdoor attacks remains entirely unexplored. To bridge this gap, we propose BadTime, the first effective backdoor attack tailored for MLTSF. BadTime can manipulate hundreds of future predictions toward a target pattern by injecting a subtle trigger. BadTime addresses two key challenges that become particularly severe in MLTSF: (i) the rapid dilution of local triggers over long horizons, and (ii) the extreme sparsity of backdoor signals under stealth constraints. To counter dilution, BadTime leverages inter-variable correlations, temporal lags, and data-driven initialization to design a distributed, lag-aware trigger whose influence persists over long-range forecasts. To overcome sparsity, it mixes label-similar and label-dissimilar poisoned samples and adopts a decoupled backdoor training objective that explicitly up-weights the sparse backdoor signal, enabling reliable learning at a poisoning rate as low as 1%. Extensive experiments show that BadTime significantly outperforms the state-of-the-art backdoor attack on time series forecasting by extending the attackable horizon from at most 12 timesteps to 720 timesteps (a 60x improvement) and reducing the MAE on target variables by 57% on average, while changing the average clean forecasting error by less than 1%. ## MEL: Coordinate-Preserving EEG Tokenization for fMRI Translation DevFeed: [MEL: Coordinate-Preserving EEG Tokenization for fMRI Translation]() Original publisher: [Read original article]() Author: Xiangyu Liu, Zeting Yan, Zhitong Yin, Boyang Li, Xi Zhang Published: 2026-10-28T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [mlp](), [fourier transform](), [reproducibility]() Tags: [aggregation](), [bootstrap](), [canonical](), [capacity](), [channel](), [confidence](), [confidence-interval](), [correlation](), [cs-lg](), [eeg](), [external](), [features](), [global](), [history](), [improvements](), [input](), [interface](), [lag](), [measurement](), [tokenization](), [translation]() ### AI overview MEL represents each preceding 16-second EEG history as calibrated lag-channel-frequency measurements and compares four equal-dimensional inputs with matched Ridge and MLP readouts. In 29 resting-state scans from 22 participants, Morlet-Ridge reached mean scan-wise correlation 0.508 across six regional targets and the global signal, versus 0.432 for canonical Fourier features; paired improvement was 0.069 (95% bootstrap CI 0.037-0.101), with gains for 19 participants. Morlet MLP reached 0.491. Shared coordinate permutations preserved prediction after refitting, while sample-wise permutations degraded it. Generalization was weak: canonical Fourier regression scored 0.377 under participant exclusion, and external unseen-participant correlation was 0.026. The authors conclude spectral estimation mattered without added nonlinear readout capacity, but call for confirmation on untouched participant-separated data with split-aware preprocessing. ### Source excerpt arXiv:2608.29304v2 Announce Type: replace Abstract: Background and Objective: Predicting regional blood-oxygen-level-dependent (BOLD) activity from EEG provides a computational approach to studying electrophysiological-hemodynamic associations. However, changes in input representation, readout architecture, and ensembling can obscure the source of a prediction gain. We investigate the contribution of target-relative spectral estimation under matched readout conditions. Methods: MEL converts each preceding 16-second EEG history into calibrated lag-channel-frequency measurements. Four equal-dimensional representations are compared using matched Ridge and multilayer perceptron (MLP) readouts. The retrospective analysis includes 29 resting-state scans from 22 participants, participant-paired uncertainty estimates, aggregation and correspondence controls, leave-one-participant-out regression, and refitted models on an external Oddball cohort. Results: Morlet-Ridge achieved a mean scan-wise correlation of 0.508 across six regional targets and the global signal, compared with 0.432 for canonical Fourier features. The participant-paired mean difference was 0.069 (95% bootstrap confidence interval, 0.037-0.101), with improvements in 19 of 22 participants. The Morlet MLP prediction ensemble reached 0.491. Shared coordinate permutations preserved prediction after refitting, whereas sample-wise permutations degraded it. Canonical Fourier regression reached 0.377 under participant exclusion; external unseen-participant correlation was 0.026. Conclusions: Spectral estimation materially affected regional BOLD prediction without requiring additional nonlinear readout capacity. MEL provides an explicit measurement interface for controlled comparisons, but weak external results constrain its generalizability. Confirmation requires untouched participant-separated data and split-aware preprocessing. ## Cautious Conflict-Free Inverse Gradient Methods DevFeed: [Cautious Conflict-Free Inverse Gradient Methods]() Original publisher: [Read original article]() Author: Damian Hattler, Qiang Liu, Nils Thuerey Published: 2026-10-26T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Neural Network](), [Artificial Intelligence]() Tags: [adaptive](), [algorithm](), [benchmarks](), [cs-lg](), [flow](), [gradient](), [lag](), [loss-functions](), [machine-learning](), [masking](), [optimization](), [partial-differential-equations](), [performance](), [physics-informed](), [vulnerability](), [weight]() ### AI overview In PINN training, ConFIG creates a conflict-free direction for multiple task gradients, but an adaptive optimizer such as Adam can alter it so the resulting weight update is no longer conflict-free. Cautious ConFIG adds a three-line extension that empirically restores the guarantee before the update. The abstract reports consistent improvements over standard ConFIG on 1D viscous Burgers', 1D nonlinear Schrödinger, and 2D Kovasznay flow benchmarks. A toy system is also used to examine cautious masking's stabilization and its response to conflict severity, task count, and optimizer lag. ### Source excerpt arXiv:2610.31855v1 Announce Type: new Abstract: Partial differential equations (PDEs) are indispensable tools for modeling complex phenomena in physics and engineering. Physics-Informed Neural Networks (PINNs) have emerged as a powerful alternative to traditional numerical solvers by directly incorporating physical constraints into neural network loss functions. Despite their advantages, PINN training requires multi-objective optimization, which often leads to significant gradient conflicts among boundary, initial, and residual loss terms. The Conflict-Free Inverse Gradient (ConFIG) method addresses these conflicts; however, integrating ConFIG with adaptive optimizers such as Adam introduces a critical vulnerability: the conflict-free direction generated by ConFIG is further processed by the optimizer, and the resulting weight update may no longer be conflict-free with respect to the task gradients. To address this issue, Cautious ConFIG is proposed as a concise three-line extension to the ConFIG algorithm that empirically restores the conflict-freeness guarantee prior to the weight update. The method is evaluated on three PINN benchmarks: the 1D Viscous Burgers' equation, the 1D Nonlinear Schr\"odinger equation, and the 2D Kovasznay flow. Experimental results demonstrate consistent performance improvements over the standard ConFIG baseline. Additionally, the stabilizing effect of cautious masking and its reaction to conflict severity, task number, and optimizer lag are investigated using a simple toy system. ## Kyutai CADRE: Video Autoencoders That Cut World Model Lag Below 230 ms DevFeed: [Kyutai CADRE: Video Autoencoders That Cut World Model Lag Below 230 ms]() Original publisher: [Read original article]() Author: Michal Sutter Published: 2026-10-23T23:26:53Z Content type: article Language: en Sources: [MarkTechPost]() Topics: [World models](), [diffusion-transformers](), [Open Source]() Tags: [agentic-ai](), [ai-infrastructure](), [ai-paper-summary](), [ai-shorts](), [applications](), [artificial-intelligence](), [autoencoders](), [diffusion](), [editors-pick](), [fps](), [interactive](), [kyutai](), [lag](), [language-model](), [large-language-model](), [machine-learning](), [model](), [new-releases](), [open](), [staff](), [tech-news](), [technology](), [video](), [world-model]() ### AI overview CADRE targets action lag in video world models by using causal diffusion autoencoders that stream decoded chunks in one or two steps. In Kyutai's evaluation, CADRE world models ran at 20 fps on one H100 and reacted in under 230 ms; a 1-step configuration reached 185 ms and 23.4 fps. Its seven variants change the encoder bottleneck's temporal and spatial compression. Results suggest moving-camera tasks favor less temporal compression, while coarse spatial tokens can hurt motion quality before image quality. CADRE performs well across many reported comparisons, but does not win every task: Cosmos scored better on robot manipulation FVD, and Wan 2.2 had much better reconstruction PSNR. Long temporal latents also showed chunk-boundary seams, and transfer to generalist world models remains unknown. The code is MIT licensed; the weights are CC BY-NC 4.0 and exclude commercial use. ### Source excerpt Kyutai released CADRE, a family of 7 open causal diffusion video autoencoders for interactive world models. World models built on CADRE run at 20 fps on 1 H100 and react in under 230 ms. The post Kyutai CADRE: Video Autoencoders That Cut World Model Lag Below 230 ms appeared first on MarkTechPost. ## Hunting Replication Lag: A Queue-by-Queue Method for Diagnosing Oracle GoldenGate DevFeed: [Hunting Replication Lag: A Queue-by-Queue Method for Diagnosing Oracle GoldenGate]() Original publisher: [Read original article]() Author: Zalak Shrof Published: 2026-10-23T20:53:46Z Content type: article Language: en Sources: [HackerNoon]() Topics: [Oracle Database](), [queueing theory](), [Parallelism]() Tags: [apply](), [backlog](), [benchmark](), [capacity](), [capture](), [checkpoint](), [commit](), [constraint](), [data-replication](), [database](), [distributed-systems](), [lag](), [oracle](), [oracle-database](), [oracle-goldengate](), [oraclegoldengate](), [parallelism](), [queue](), [replicat](), [replication](), [replication-lag]() ### AI overview GoldenGate lag is the buildup of work in a pipeline of queues, so adding Replicat appliers helps only when apply capacity is the constraint. The article recommends measuring each hop and using target heartbeat tables for end-to-end lag; process LAG reports the age of the record being processed and can look healthy while newer data is delayed upstream. Its model illustrates how large-transaction barriers, hot-row dependencies, unusable target indexes, or network throttling can limit applier gains. A missing target index can leave scans in place even when KEYCOLS identifies rows, and per-key ordering caps throughput on hot rows. After a bottleneck clears, backlog shifts downstream, which then determines drain time; in active-active replication, the lag also widens the conflict window. The model is illustrative, omits applier contention and other behaviors, and overstates the benefit of splitting transactions by assigning pieces unique keys. ### Source excerpt When replication falls behind, the reflex is to raise Replicat parallelism, which works only when apply is the binding constraint. ## Laravel Read Replicas: Prevent Stale Reads After a Write DevFeed: [Laravel Read Replicas: Prevent Stale Reads After a Write]() Original publisher: [Read original article]() Author: Paul Redmond Published: 2026-10-22T23:20:21Z Content type: tutorial Language: en Sources: [Laravel]() Topics: [read-replica](), [Laravel](), [middleware](), [consistency](), [Eloquent ORM]() Tags: [database](), [lag](), [laravel](), [middleware](), [news](), [prevent](), [read-replicas](), [replica](), [replica-lag](), [write]() ### AI overview Laravel 13.38 adds `PreventStaleDatabaseReads` to address replica lag across separate requests: after a request modifies records, the middleware stores a session timestamp, then routes reads to the write server on subsequent requests for a configurable window (two seconds by default). Set the window to cover the longest lag seen in replica metrics; a shorter window can still expose stale data. This differs from the `sticky` connection option, which only keeps reads on the writer for the remainder of the same request. For code that needs fresh reads regardless of the user's recent writes, `DB::usingWritePdo()` routes all reads inside a callback to the writer, including relationship queries; the connection returns to its prior setting when the callback finishes or throws. Queries in a transaction already use the writer. The article also notes that middleware tests should avoid `RefreshDatabase`, whose transaction would make queries use the writer even without the middleware. ### Source excerpt Laravel's PreventStaleDatabaseReads middleware keeps a user's reads on the write database after they save, so read replica lag doesn't show them old data. The post Laravel Read Replicas: Prevent Stale Reads After a Write appeared first on Laravel News. Join the Laravel Newsletter to get Laravel articles like this directly in your inbox. ## Troubleshooting Data Mismatches with tablediff on Amazon RDS for SQL Server DevFeed: [Troubleshooting Data Mismatches with tablediff on Amazon RDS for SQL Server]() Original publisher: [Read original article]() Author: Kanchan Bhattacharyya Published: 2026-10-22T16:33:04Z Content type: article Language: en Sources: [AWS Database Blog]() Topics: [RDS for SQL Server](), [read-replica](), [sql-server](), [Amazon RDS](), [SQL](), [observability](), [Disaster Recovery](), [audit](), [Amazon S3]() Tags: [2022](), [2025](), [advanced-300](), [amazon-cloudwatch](), [amazon-ec2](), [amazon-elastic-compute-cloud-amazon-ec2](), [amazon-rds](), [amazon-simple-storage-service-amazon-s3](), [asynchronous](), [audit](), [best-practices](), [how-to](), [lag](), [rds-for-sql-server](), [read-replica](), [replication](), [scripts](), [sql-server](), [t-sql](), [technical-how-to](), [troubleshooting](), [utility]() ### AI overview After an Amazon RDS for SQL Server read replica catches up, tablediff can provide independent row-level parity evidence that replication metrics alone do not: compare the primary and replica only after `ReplicaLag`, `redo_queue_size`, and `log_send_queue_size` are all zero. Run it from a Windows EC2 client with access to both endpoints; target tables need a clustered primary key or unique index. It reports source-only, destination-only, and column-mismatched rows, and can generate T-SQL fixes. The replica is read-only, so remediation requires promotion, which is irreversible; regenerate the fix script after promotion and wrap its bare DML statements in an explicit transaction with a validation gate. Re-run tablediff to confirm zero differences. A fresh rebuild may be preferable for extensive or structural divergence. Comparisons scan whole tables and can give misleading results during active writes or redo, so run during a quiet period after queues drain. ### Source excerpt Learn how to use the SQL Server tablediff utility to validate an Amazon RDS for SQL Server read replica against its primary after a replication lag event, detect and classify row-level data mismatches, and apply auditable T-SQL fix scripts on a promoted replica instance. ## Delay-Guided Attention for Quality Prediction in Multistage Manufacturing DevFeed: [Delay-Guided Attention for Quality Prediction in Multistage Manufacturing]() Original publisher: [Read original article]() Author: Jiarui Zhang, Zhaoyang Guan, Zhixuan Xia, Naichen Shi Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Time-sequences](), [Representation learning](), [Transformer]() Tags: [ablation](), [alignment](), [attention](), [benchmarks](), [cs-lg](), [delay](), [evaluations](), [forecasting](), [fusion](), [lag](), [manufacturing](), [quality](), [temporal]() ### AI overview DAMF improves final quality prediction by estimating time-varying delays between adjacent manufacturing stages and using those estimates to guide cross-stage attention. Its lag predictor compares downstream representations with candidate lagged upstream representations; an occurrence-magnitude factorization produces sample- and stage-pair-specific distributions over discrete lag states without plant-specified delays at inference. DelayAlignment uses these distributions as soft temporal priors, while gated residual fusion regulates information flow. For interpretation, segment-wise Viterbi decoding produces temporally coherent lag paths; forecasting retains soft distributions for alignment. The abstract reports better quality prediction than compared baselines on the LiquidSugar dataset and more accurate, stable lag recovery on two controlled-delay benchmarks. Ablations support the contributions of lag prediction, delay-guided attention, and structured temporal decoding. ### Source excerpt arXiv:2610.29220v1 Announce Type: new Abstract: Quality prediction in multistage manufacturing is challenging when inter-stage lags vary over time, as measurements taken at the same timestamp may reflect different process states. Existing approaches often preserve production order but fuse timestamp-aligned features or model delayed dependencies implicitly, without identifying the upstream history relevant to each downstream observation. We propose Delay-Guided Attention for Multi-stage Forecasting (DAMF), which explicitly estimates inter-stage lag distributions and uses them to guide cross-stage information transfer for final quality prediction. For each adjacent stage pair, a lag predictor compares the current downstream representation with candidate lagged upstream representations. An occurrence-magnitude factorization separately estimates the probability of a positive lag and its conditional magnitude, yielding a sample- and stage-pair-specific distribution over discrete lag states without requiring plant-specified delays at inference. DelayAlignment incorporates this soft distribution as a temporal prior into differentiable cross-stage attention, while gated residual fusion regulates information propagation across production stages. For interpretation, segment-wise Viterbi decoding converts pointwise lag distributions into temporally coherent lag paths, whereas forecasting retains the soft distributions for alignment. On the real-world LiquidSugar dataset, DAMF improves final quality prediction over the compared baselines. Evaluations on two controlled-delay benchmarks show more accurate and stable lag recovery, and ablation studies support the contributions of lag prediction, delay-guided attention, and structured temporal decoding. ## Discount Yourself: Self-inclusive Freshness-weighted Gossip for Decentralised Learning under Client Outages DevFeed: [Discount Yourself: Self-inclusive Freshness-weighted Gossip for Decentralised Learning under Client Outages]() Original publisher: [Read original article]() Author: Qiyuan Wang, Ke Xiao, Christos Anagnostopoulos Published: 2026-10-21T04:00:00Z Content type: article Language: en Sources: [arXiv: Distributed Computing](), [arXiv: Machine Learning]() Topics: [Federated Learning]() Tags: [aggregation](), [cs-dc](), [cs-lg](), [federated-learning](), [lag](), [outages](), [smartphone-availability]() ### AI overview In decentralised federated learning, a returning device's own model can be the stalest candidate after an outage, but existing rules often discount only incoming neighbour models. Self-inclusive Freshness-weighted Gossip (SFG) addresses this by discounting the receiver alongside neighbours, measuring inherited lag rather than model age, and applying a steep discount because lag persists across rounds. Its effective-lag measure adds one scalar per message; the proposed rules are described as matching decentralised federated averaging's communication and computation, with guarantees for the stale receiver. Across three image benchmarks and eight outage processes, including smartphone availability traces, SFG-L and SFG-Cut consistently outperform decentralised federated averaging and recent age-of-information rules. The supplied excerpt does not provide further experimental details. ### Source excerpt arXiv:2610.28545v1 Announce Type: cross Abstract: Decentralised federated learning (DFL) trains personalised models on devices that average with their neighbours rather than with a central server. Real devices, however, are often offline for hours or days, and a client returning from an outage can recover only by aggregating the models its neighbourhood holds. We observe that the stalest of these models is usually the client's own, a case that existing staleness-aware rules, which discount only the models a client receives, do not address. We formalise aggregation after an outage as estimating the state of a fully available network and derive three principles. First, the receiver must be discounted like any neighbour. Every rule that weights it at least as much as each neighbour has an error that grows without bound with the outage, whereas discounting all candidates by relative staleness keeps the error bounded. Second, staleness must be measured as inherited lag rather than age, since local training resets a model's age but not the lag it absorbed from stale neighbours. We introduce the effective lag, which tracks this quantity at the cost of one scalar per message. Third, the discount must be steep, because absorbed lag persists in every later round. These principles define Self-inclusive Freshness-weighted Gossip (SFG), a family of drop-in aggregation rules with the communication and computation of decentralised federated averaging, whose guarantees hold for the stale receiver itself. On three image benchmarks and eight outage processes, including real smartphone availability traces, the lag-based members SFG-L and SFG-Cut consistently outperform decentralised federated averaging and recent age-of-information rules, and ablations show that each principle is necessary. The analysis also explains why tuned limited merging from gossip learning is effective. ## KDE Plasma 6.8 Ready For Release Next Week, Sees Last Minute NVIDIA Fix DevFeed: [KDE Plasma 6.8 Ready For Release Next Week, Sees Last Minute NVIDIA Fix]() Original publisher: [Read original article]() Author: Michael Larabel Published: 2026-10-10T10:00:51Z Content type: news Language: en Sources: [Phoronix]() Topics: [CSS Framework](), [MTU mismatch]() Tags: [desktop-linux](), [frameworks](), [kde](), [keyboard](), [lag](), [latest](), [linux-benchmarking](), [linux-hardware-benchmarks](), [linux-hardware-reviews](), [linux-how-to](), [linux-performance](), [linux-server-benchmarks](), [main-thread](), [nvidia](), [open-source-graphics](), [opengl](), [phoronix](), [phoronix-test-suite](), [plasma](), [qt](), [qt-quick](), [release](), [stuttering](), [ubuntu-benchmarks](), [ubuntu-hardware](), [up](), [virtual-desktop](), [work]() ### AI overview KDE Plasma 6.8 is expected to debut next week. Its final fixes include improved KRunner keyboard navigation, more reliable automatic updates for Flatpak apps, and a change addressing lag and stuttering when tiling windows with certain NVIDIA GPUs. The report also notes KDE Frameworks and upcoming Plasma 6.9 changes. ### Source excerpt KDE Plasma 6.8 is gearing up for its official debut next week in coinciding with the 30th anniversary since the KDE project began... ## Valve Lands VK\_AMD\_anti\_lag Support For AMD RADV Driver In Mesa 26.3 DevFeed: [Valve Lands VK\_AMD\_anti\_lag Support For AMD RADV Driver In Mesa 26.3]() Original publisher: [Read original article]() Author: Michael Larabel Published: 2026-10-09T00:23:33Z Content type: news Language: en Sources: [Phoronix]() Topics: [firmware-update](), [rocm]() Tags: [amd](), [desktop-linux](), [gaming](), [gpu](), [lag](), [latency](), [linux](), [linux-benchmarking](), [linux-hardware-benchmarks](), [linux-hardware-reviews](), [linux-how-to](), [linux-performance](), [linux-server-benchmarks](), [low-latency](), [open-source-graphics](), [phoronix](), [phoronix-test-suite](), [radeon](), [release](), [ubuntu-benchmarks](), [ubuntu-hardware](), [valve](), [vk](), [vulkan]() ### AI overview Valve engineers merged VK_AMD_anti_lag support for the AMD RADV driver into Mesa 26.3. The Vulkan extension paces CPU work to reduce latency between input events and screen updates, and the new Mesa implementation uses a shared frame-pacing utility that may also support NVIDIA's VK_NV_low_latency_2 extension. ### Source excerpt There is yet another feature to get excited about for this quarter's Mesa 26.3 release. Months in the making, the Valve-contributed VK_AMD_anti_lag support for the Radeon RADV driver was merged today for benefiting AMD Linux gamers as a nice improvement over the former implementation... [Next page]()