# validation Published articles for validation. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Finding and using interpretable latents in a neutrino foundation model with sparse autoencoders DevFeed: [Finding and using interpretable latents in a neutrino foundation model with sparse autoencoders]() Original publisher: [Read original article]() Author: Rapha\\"el Bonnet-Guerrini, Johann Ioannou-Nikolaides, Inar Timiryasov, Vincenzo Piuri Published: 2026-11-09T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [AI Foundation Models](), [Representation learning](), [reproducibility]() Tags: [astro-ph-he](), [autoencoders](), [cs-ai](), [cs-lg](), [error](), [foundation-model](), [hep-ex](), [icecube](), [mechanistic-interpretability](), [neutrino](), [particle-physics](), [replication](), [uncertainty](), [validation]() ### AI overview The study applies sparse autoencoders to a neutrino foundation model pretrained on IceCube data and fine-tuned for direction reconstruction. Held-out tests, matched nuisance controls, and replication across independent dictionary trainings support an atlas of physical concepts in the model's representation. Causal interventions indicate that the direction head barely uses this atlas. An uncertainty head trained on the same event-level representation predicts angular reconstruction error and causally depends on atlas features related to event quality and brightness. The findings suggest interpretable latent physics can inform downstream task design. ### Source excerpt arXiv:2608.26090v2 Announce Type: replace-cross Abstract: We present a first application of sparse-autoencoder-based mechanistic interpretability to particle physics. Studying a neutrino foundation model pretrained on IceCube data and fine-tuned for direction reconstruction, we identify a validated atlas of physical concepts in the model representation, using a strict validation protocol consisting of held-out tests, matched nuisance controls, and replication across independent dictionary trainings. Causal interventions show that the direction head barely draws on this atlas. Motivated by this underused information, we train an uncertainty head on the same event-level representation to predict the model's angular reconstruction error. Unlike the direction head, it depends causally on quality and brightness features from the atlas. These results suggest that mechanistic interpretability can reveal learned latent physics encoded within a model's internal representation and help design downstream tasks that exploit it. ## Reliable Prerequisite-Aware Learning Path Recommendation under Uncertain Knowledge Structures: Long-Horizon Evaluation and Cross-Dataset Validation DevFeed: [Reliable Prerequisite-Aware Learning Path Recommendation under Uncertain Knowledge Structures: Long-Horizon Evaluation and Cross-Dataset Validation]() Original publisher: [Read original article]() Author: M. M. Golam Hafiz, A. K. M. Shahriar Emtiaz, Safia Akter Simi, Md. Rakibul Islam, M. F. Mridha Published: 2026-11-09T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [recommendation systems](), [Machine learning]() Tags: [autoregressive](), [cs-ai](), [cs-lg](), [dataset](), [evaluation](), [graph](), [hierarchy](), [horizon](), [knowledge](), [learning](), [recommendation](), [validation]() ### AI overview The study finds that strong next-resource ranking does not establish reliable long-horizon learning paths. In a frozen MOOCCubeX evaluation, XGBoost pairwise had the highest observed MRR (0.9048), above the prerequisite-aware graph model (0.7707); a development ablation also found the no-graph condition strongest for immediate ranking, which depended heavily on permitted title and path similarity. Autoregressive evaluation exposed path properties missed by one-step ranking. On Junyi, which uses hierarchy metadata rather than an explicit prerequisite graph and is not zero-shot transfer, Markov had the highest observed MRR (0.4453) versus 0.3954 for the graph model, with differences across ranking metrics. The authors argue that immediate accuracy, prerequisite reliability, and long-horizon path quality are complementary evaluation dimensions. The primary MOOCCubeX task is course-prefix-conditioned, not learner-personalized. ### Source excerpt arXiv:2611.09345v1 Announce Type: new Abstract: Learning-path recommendation often depends on knowledge and prerequisite structures whose reliability is uncertain, while high accuracy in predicting the next resource does not necessarily establish reliable long-horizon path behaviour. This study develops a reliability-oriented prerequisite-aware framework that jointly examines immediate next-resource ranking, uncertain prerequisite structures, hard, soft and hybrid constraint semantics, graph robustness, and autoregressive path generation under course-level confirmatory evaluation and independent cross-dataset validation. Development used 3,187 MOOCCubeX courses, followed by a frozen confirmatory evaluation on 541 held-out courses and 37,747 transitions. The primary MOOCCubeX task is structural and course-prefix-conditioned rather than learner-personalised. Junyi provides independent structure-/hierarchy-aware external validation, uses hierarchy metadata rather than an explicit prerequisite graph and is not a zero-shot transfer setting. On the frozen MOOCCubeX evaluation, XGBoost pairwise was the strongest observed immediate ranker, with mean reciprocal rank (MRR) of 0.9048, whereas the proposed prerequisite-aware graph model achieved 0.7707. In the controlled graph-variant development ablation, the no-graph condition produced the strongest immediate ranking result, and immediate prediction showed strong dependence on permitted title/path-similarity information. Autoregressive evaluation exposed path properties not captured by one-step ranking. On Junyi, Markov achieved the strongest observed MRR of 0.4453, compared with 0.3954 for the proposed graph model, with metric-specific differences across ranking measures. Overall, immediate predictive accuracy, prerequisite-structure reliability, and long-horizon path quality should be treated as complementary rather than interchangeable evaluation dimensions in educational recommendation. ## CVE-2026-123596 - Hickory DNS through 0.26.0-alpha.1 DNSSEC Validation Bypass via Ancestor-Delegation NSEC DevFeed: [CVE-2026-123596 - Hickory DNS through 0.26.0-alpha.1 DNSSEC Validation Bypass via Ancestor-Delegation NSEC]() Original publisher: [Read original article]() Author: CVEFEED.io; vulncheck.com Published: 2026-11-08T02:16:54Z Content type: article Language: en Sources: [Latest High/Critical Vulnerabilitiy Feed]() Topics: [DNSSEC](), [DNS security](), [Domain Name System (DNS)](), [TTL in DNS]() Tags: [bypass](), [cve-2026-123596](), [delegation](), [dns](), [dnssec](), [validation]() ### AI overview Hickory DNS has a DNSSEC validation flaw in the validating recursor's NSEC handler: it accepts parent-side ancestor-delegation NSEC records as proof that names below a zone cut do not exist. An on-path attacker, rogue parent nameserver, or attacker who wins a port and transaction-ID race can use a forged NXDOMAIN response containing genuinely signed parent NSEC records to make existing child-zone names appear absent; that denial can remain cached for the negative TTL. The excerpt identifies hickory-proto through 0.26.0-alpha.1 and hickory-net 0.26.0-beta.1 before 0.26.2 as affected. It gives a severity score of 8.8 (High), but provides no fix or mitigation details. ### Source excerpt CVE ID :CVE-2026-123596 Published : Nov. 8, 2026, 2:16 a.m. | 47 minutes ago Description :Hickory DNS hickory-proto through 0.26.0-alpha.1 and hickory-net 0.26.0-beta.1 before 0.26.2 contain a DNSSEC validation flaw in the validating recursor's NSEC handler that accepts parent-side ancestor-delegation NSEC records as proof of nonexistence for names below the zone cut. An on-path attacker, rogue parent nameserver, or attacker winning a port and TXID race can deliver one forged NXDOMAIN response built from genuinely signed parent NSEC records, causing authenticated denial of existing child-zone names cached for the negative TTL. Severity: 8.8 | HIGH Visit the link for more details, such as CVSS details, affected products, timeline, and more... ## When Blockchain Data Becomes AI Context, Stale State Becomes a Security Risk DevFeed: [When Blockchain Data Becomes AI Context, Stale State Becomes a Security Risk]() Original publisher: [Read original article]() Author: Atul Kumar Published: 2026-11-06T08:29:55Z Content type: article Language: en Sources: [HackerNoon]() Topics: [Blockchain](), [context](), [ai security](), [Data Quality](), [consistency](), [AI Architecture](), [Agent Harness](), [state-management](), [agentic workflows](), [observability](), [AI-Ready Data](), [AI Integration Strategies]() Tags: [account](), [account-balance](), [accounting](), [actions](), [agent](), [agent-runtimes](), [agents](), [ai](), [ai-agent](), [ai-agents](), [ai-engineering](), [ai-systems](), [artificial-intelligence](), [blockchain](), [context](), [context-engineering](), [cybersecurity](), [data](), [distributed-systems](), [execution](), [risk](), [security](), [smart-contracts](), [state](), [validation](), [web3]() ### AI overview Blockchain data can be accurate when retrieved yet unsafe by the time an AI agent acts: balances, liquidity, or lending positions may change, and RPC reads, indexers, events, and price feeds may describe different states. The article argues that retrieval success or a timestamp alone cannot establish decision validity. It recommends tracking each observation's source and block reference, checking freshness and consistency against the action's risk, and refreshing or rejecting context when required. Proposed transactions should carry their critical assumptions and be revalidated before signing; after submission, the system should verify transaction status and resulting state rather than assume success or retry blindly. Historical memory should preserve what was observed without standing in for current state. Stronger checks improve safety but add latency and infrastructure cost, so requirements should vary by task, with consequential actions requiring stronger evidence. The article also recommends tracing observations, policy checks, authorization, and confirmed outcomes to explain decisions. ### Source excerpt Stale blockchain data can lead AI agents to make unsafe decisions. Learn how context validation, state tracking, and execution checks improve reliability. ## EmboTeam: Asymmetric State Grounding for Heterogeneous Multi-Robot Collaboration DevFeed: [EmboTeam: Asymmetric State Grounding for Heterogeneous Multi-Robot Collaboration]() Original publisher: [Read original article]() Author: Haishan Zeng, Mengna Wang, Peng Li Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Machine Learning]() Topics: [Multi Agent Systems](), [synchronization](), [Large Language Model](), [Multi-step AI processes]() Tags: [behavior](), [collaboration](), [cs-ai](), [cs-cv](), [cs-lg](), [cs-ma](), [cs-ro](), [execution](), [features](), [framework](), [generate](), [global](), [grounding](), [language-models](), [reactive](), [robot](), [state](), [structured](), [synchronization](), [tasks](), [team](), [thor](), [validation]() ### AI overview EmboTeam addresses long-horizon coordination failures in heterogeneous robot teams by combining LLM-generated state expectations with PDDL symbolic verification and a team-level parallel behavior tree. Robot-specific subtrees check preconditions, execute skills, validate effects, replan locally, and synchronize through a blackboard. On MACE-THOR, a benchmark of 42 tasks across 8 indoor layouts, it raises task success from 12% to 55% and goal condition recall from 32% to 75% compared with LaMMA-P. The supplied text is an abstract-level excerpt, so it does not detail evaluation conditions or limitations. ### Source excerpt arXiv:2601.11063v3 Announce Type: replace-cross Abstract: Heterogeneous multi-robot collaboration over long horizons requires both global task planning and robust coordination during execution. Existing plan- ners using large language models (LLMs) can generate task decompositions or symbolic plans, but often execute them as rigid action sequences, making them fragile to local state mismatches and synchronization failures. We propose Em- boTeam, a neuro-symbolic framework for asymmetric state grounding. An LLM generates structured state expectations, which are verified and coordinated through symbolic search using Planning Domain Definition Language (PDDL) and then grounded into a team-level parallel behavior tree. Robot-specific subtrees per- form precondition checking, skill execution, effect validation, local replanning, and blackboard-based synchronization. We further introduce MACE-THOR, con- taining 42 heterogeneous multi-robot tasks across 8 indoor layouts. EmboTeam improves task success rate from 12% to 55% and goal condition recall from 32% to 75% over LaMMA-P, demonstrating the effectiveness of synchronized reactive grounding for collaboration over long horizons. ## How Far Can Lightweight LLMs Go in Intent-to-SLO Translation? DevFeed: [How Far Can Lightweight LLMs Go in Intent-to-SLO Translation?]() Original publisher: [Read original article]() Author: Sabbir Ahmed, Hazem Abbas, Salimur Choudhury Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Networking]() Topics: [Large Language Model](), [Chain-of-thought](), [few-shot](), [cisco networking]() Tags: [cloud-hosted-infrastructure](), [cost](), [cs-cl](), [cs-ni](), [energy](), [evaluation](), [gpu](), [infrastructure](), [intent-based-networking](), [language-models](), [lightweight](), [multi-gpu](), [network](), [operator](), [prompting](), [qos](), [reasoning](), [robustness](), [slos](), [translation](), [validation](), [zero-shot]() ### AI overview The study evaluates ten open LLMs with fewer than 10 billion parameters on translating 716 network placement intents into service-level objectives. Best configurations reach strict F1 of 0.943 and reject up to 191 of 195 invalid intents. Few-shot reasoning limits unsupported QoS requirements to about one in a hundred responses for the strongest models, while zero-shot performance struggles; reasoning adds energy cost but pays off for weaker models. Translation without reasoning uses 0.044-0.148 Wh of GPU energy. Realistic rephrasing reduces invalid-intent rejection in 15 of 16 configurations for the best-performing models, making robust validation an unresolved challenge. ### Source excerpt arXiv:2611.07982v1 Announce Type: new Abstract: Intent-based networking (IBN) promises QoS provisioning in which the network derives the configuration from operator-stated goals. Its first step, translating a natural-language intent into enforceable service-level objectives (SLOs) grounded in a live snapshot of the network state, increasingly relies on large language models (LLMs). These large models add cost and energy to the management loop and typically require multi-GPU, cloud-hosted infrastructure. Can lightweight open LLMs take over this step dependably? We answer this question with a five-dimensional evaluation of ten LLMs with <10B parameters from five families, under four prompting setups, on 716 placement intents. We evaluate translation performance, intent validation, hallucination, energy impact, and robustness to diverse phrasing. The best configurations translate intents with a strict F1 of 0.943 and reject up to 191 of 195 invalid intents. With few-shot reasoning, the strongest models hallucinate unsupported QoS requirements in only about one response in a hundred, whereas the zero-shot setup struggles. A translated intent costs 0.044-0.148 Wh of GPU energy without reasoning, and added energy cost for reasoning pays off for the weaker models. Furthermore, we show that realistic rephrasing lowers the rejection of invalid intents in 15 of 16 configurations for the best-performing models, leaving robust intent validation as a key open challenge for reliable IBN. Our code and data are available: https://github.com/ggck43/intent2slo. ## Machine Learning for Live-Birth Prediction After IVF/ICSI From Structured Clinical Data: A Systematic Review, Meta-Analysis, and Critical Appraisal DevFeed: [Machine Learning for Live-Birth Prediction After IVF/ICSI From Structured Clinical Data: A Systematic Review, Meta-Analysis, and Critical Appraisal]() Original publisher: [Read original article]() Author: Abdul Raheem Shahzad, Muhammad Zaheer Khan, Umar Ajaib, Asad Ur Rehman, Juan Deng, Ming Zhang Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Machine learning](), [Predictive AI](), [generalization in machine learning]() Tags: [benchmark](), [cs-lg](), [ivf](), [local](), [logistic-regression](), [machine-learning](), [meta](), [prediction](), [prisma](), [review](), [risk](), [validation]() ### AI overview Across 48 studies, machine-learning models predicting live birth after IVF/ICSI showed highly variable performance, and the evidence remains weak: 47 studies had high or unclear risk of bias, only four assessed calibration slope, and none demonstrated clinical impact. Internal-only models pooled at AUC 0.814, compared with 0.756 for evaluation in a later period or another centre; this difference was not significant, heterogeneity was extreme, and overall certainty was low to very low. The only significant meta-regression moderator was an unstated decision point, associated with a pooled AUC of 0.910. Compared with logistic regression on the same data, the best machine-learning model gained 0.056 AUC. The review concludes that gains surviving validation came from staging predictions to decision points and local calibration, rather than deeper architectures. ### Source excerpt arXiv:2611.07124v1 Announce Type: new Abstract: Machine-learning models predicting live birth after in vitro fertilisation (IVF) or intracytoplasmic sperm injection (ICSI) from routine clinical data report performance from near chance to near perfect. We systematically reviewed these models under PRISMA 2020, searching OpenAlex, Scopus and Web of Science to 22 October 2026, appraised them with PROBAST+AI and TRIPOD+AI, and pooled their discrimination with random-effects meta-analysis on the logit scale. We define what fixes the meaning of a model's output (decision point, endpoint, unit of analysis, data source) and map the 48 included studies by model family, decision point and validation design. Models evaluated only internally pooled to an area under the receiver operating characteristic curve (AUC) of 0.814, against 0.756 (95% prediction interval 0.546-0.889) for models evaluated on a later period or another centre, and 0.764 for studies not at high risk of bias; neither difference was significant, and heterogeneity was extreme. Models whose decision point was not stated pooled to 0.910, the only significant moderator in meta-regression. Against logistic regression fitted to the same data, the best machine-learning model gained a pooled 0.056 AUC (95% CI 0.017-0.090). Of 48 studies, 47 were at high or unclear risk of bias, only four assessed the calibration slope, and no model has evidence of clinical impact. The certainty of the pooled evidence is low to very low; the gains that survive validation came from staging predictions to decision points and local calibration, not deeper architectures. We provide a TRIPOD+AI-mapped checklist, a benchmark specification and a research agenda. ## Interpretable Multimodal MRI-Clinical Fusion for Functional Outcome Prediction After Acute Ischemic Stroke DevFeed: [Interpretable Multimodal MRI-Clinical Fusion for Functional Outcome Prediction After Acute Ischemic Stroke]() Original publisher: [Read original article]() Author: Sina Raeisadigh, Henning M\\"uller, Abderrahmane Hedjoudje Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Machine Learning]() Topics: [Medical imaging](), [Embeddings](), [Representation learning](), [Data analysis]() Tags: [activation](), [adc](), [ais](), [biomarkers](), [cs-ai](), [cs-cv](), [cs-lg](), [eess-iv](), [embeddings](), [feature](), [functional](), [fusion](), [information](), [linear](), [machine-learning](), [maps](), [modeling](), [models](), [multimodal](), [prediction](), [principal-component-analysis](), [support-vector-machines](), [validation](), [variables]() ### AI overview In 66 patients with acute ischemic stroke, clinical-only models already predicted three-month functional outcome well (AUC 0.899 ± 0.151), while MRI-only models had weaker discrimination (about 0.67-0.73). The study combined day-1 ADC, DWI, and FLAIR MRI embeddings with clinical variables and ADC-derived lesion biomarkers, using fold-isolated PCA and linear SVMs evaluated by eight-fold stratified group cross-validation. FLAIR-clinical fusion had the highest mean AUC (0.948 ± 0.061), with ADC-clinical fusion close behind (0.941 ± 0.094); adding all three MRI modalities did not consistently improve on single-modality clinical fusion. Feature-importance analysis highlighted age, NIHSS subscores, ADC lesion burden, and learned FLAIR features. These results suggest complementary imaging information, but are cross-validated findings from a small cohort. ### Source excerpt arXiv:2611.08154v1 Announce Type: cross Abstract: This study evaluates the added prognostic value of day-1 multimodal MRI for predicting three-month functional outcome after acute ischemic stroke (AIS). Sixty-six AIS patients with ADC, DWI, FLAIR, and clinical data were analyzed. Three-dimensional pretrained ResNet embeddings were extracted from each MRI modality, reduced using fold-isolated principal component analysis, and combined with structured clinical variables and ADC-derived lesion biomarkers. Linear support vector machines were evaluated using eight-fold stratified group cross-validation across MRI modality combinations. Clinical-only modeling achieved strong performance (AUC = $0.899 \pm 0.151$), while MRI-only models showed limited discrimination (AUC $\approx$ $0.67$-$0.73$). Among multimodal configurations, FLAIR-clinical fusion achieved the highest mean cross-validated performance (AUC = $0.948 \pm 0.061$), followed by ADC-clinical fusion (AUC = $0.941 \pm 0.094$). Full ADC-DWI-FLAIR fusion did not consistently outperform single-modality clinical fusion. Feature-importance analysis identified age, NIHSS subscores, ADC lesion burden, and learned FLAIR representations as key contributors, while qualitative FLAIR activation maps showed responses over lesion-like regions. These findings suggest that compact day-1 FLAIR and ADC representations provide complementary prognostic information beyond clinical variables and support interpretable, data-efficient outcome prediction in AIS. ## A latent ODE approach to full-cycle biventricular cardiac motion modeling from cine MRI DevFeed: [A latent ODE approach to full-cycle biventricular cardiac motion modeling from cine MRI]() Original publisher: [Read original article]() Author: David Br\\"uggemann, Ekaterina Krymova, Firat \\"Ozdemir, Jochen von Spiczak, Sebastian Kozerke, Samia Mora, Robert Manka, Mathieu Salzmann, Olga V. Demler Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Machine learning](), [Neural Network](), [3D](), [Image processing]() Tags: [cs-ai](), [cs-cv](), [evaluation](), [failure](), [graph](), [imaging](), [modeling](), [motion](), [neural](), [performance](), [risk](), [validation]() ### AI overview The study models biventricular anatomy and full-cycle cine MRI motion as a continuous latent trajectory, using heart-rate-aware neural ODE dynamics and a graph-based mesh autoencoder to reconstruct 3D+t motion. In 72,386 UK Biobank participants without baseline cardiovascular disease (367 incident heart failure events), adding the latent score to refitted pooled cohort equations raised the held-out stratified C-index from 0.704 to 0.785; seven established cardiac markers reached 0.764. The model also offered the best reported trade-off among tested graph/non-graph and ODE/non-ODE approaches for reconstruction fidelity, generative realism, and prognostic performance. The authors say external validation in more representative patient cohorts is needed before clinical risk-prediction use. ### Source excerpt arXiv:2606.26718v2 Announce Type: replace Abstract: Cardiac magnetic resonance imaging (CMR) captures rich spatiotemporal information about ventricular structure and motion, but conventional risk models use only a few image-derived indices from selected cardiac phases. We present a latent dynamical model that encodes bi-ventricular anatomy and full-cycle cine motion as a continuous latent trajectory, using heart-rate-aware neural ordinary differential equation (ODE) dynamics and a graph-based mesh autoencoder to reconstruct anatomically consistent 3D+t ventricular motion. A covariate-conditioned prior defines the expected end-diastolic latent state, and a Cox proportional hazards model tests whether deviations from this prior predict incident heart failure. We studied 72,386 UK Biobank participants without baseline cardiovascular disease, including 367 incident heart failure events. In a held-out evaluation subset, adding the latent score to refitted pooled cohort equations improved the stratified C-index from 0.704 to 0.785, compared with 0.764 for seven established cardiac markers. Compared with non-graph and non-ODE approaches, the proposed model gave the best trade-off between reconstruction fidelity, generative realism, and downstream prognostic performance. These results suggest that continuous full-cycle modeling of ventricular motion provides informative cardiac phenotypes beyond conventional CMR summaries, while external validation in more representative patient cohorts is required before clinical risk-prediction use. ## Decorators for Timing, Retries, Caching, and Validation in Python DevFeed: [Decorators for Timing, Retries, Caching, and Validation in Python]() Original publisher: [Read original article]() Author: Nahla Davies Published: 2026-11-05T13:25:53Z Content type: tutorial Language: en Sources: [MachineLearningMastery.com]() Topics: [Python](), [Learning](), [Pydantic]() Tags: [caching](), [decorators](), [practical-python](), [python](), [retries](), [timing](), [validation]() ### AI overview Python decorators wrap a function at its call boundary, letting shared behavior live outside the function while preserving its contract. The article explains the closure pattern, `@` reassignment and stacking order, and recommends `functools.wraps` so names, signatures and `__wrapped__` remain useful. Its examples use `try/finally` with `time.perf_counter()` for timing; a configurable retry factory that catches only selected exceptions and re-raises on the final attempt; `functools.lru_cache` for bounded memoization of pure calls with hashable arguments; and `inspect.signature().bind()` plus predicates for input validation. It cautions that sleeping retries occupy a thread and may be unsafe for non-idempotent operations, unbounded caches can retain memory, and decorators are a poor fit for complex data schemas or business logic. Keep wrappers small, preserve returns and exceptions, and order stacked decorators deliberately. ### Source excerpt In this article, you will learn how Python decorators work under the hood and how to build reliable ones for timing, retries, caching, and input... ## Hybrid LLM-Deterministic Pipelines for Enterprise-Scale Code Migration at Bloomberg DevFeed: [Hybrid LLM-Deterministic Pipelines for Enterprise-Scale Code Migration at Bloomberg]() Original publisher: [Read original article]() Author: Mario Corchero, David Williams, Laszlo Kiss Kollar, Haowen Liu, Shivam Bairoliya, Gala Malbasic, Serkan Kirbas, Federica Sarro Published: 2026-11-05T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Software Engineering]() Topics: [AI Integration Strategies](), [Large Language Model](), [CI/CD](), [Pull Request](), [observability pipelines]() Tags: [bloomberg](), [build-system](), [c-plus-plus](), [code](), [cs-ai](), [cs-se](), [enterprise](), [hybrid](), [llm](), [migration](), [packaging](), [pipelines](), [pull-requests](), [python](), [repositories](), [runtime](), [scale](), [scaling](), [telemetry](), [test](), [testing](), [validation]() ### AI overview Bloomberg's Automated Uplift initiative applied hybrid LLM and deterministic pipelines to two code migration campaigns: Python packaging across about 1,900 candidate repositories and a C++ build system migration across about 6,700. They generated 2,333 pull requests, with 729 successful migrations. The pipeline maximized deterministic processing, limiting the LLM to bounded file transformations using precomputed context, and incorporated production telemetry such as execution logs and usage traces. The abstract identifies review, testing, and deployment by repository owners as the main bottleneck, rather than migration generation. Pipeline verification before opening pull requests reduced these costs in the C++ campaign, where compilation and tests could be checked; fixes also carried over more readily across the more uniform repository population. The evidence here is the paper's abstract. ### Source excerpt arXiv:2611.05457v1 Announce Type: new Abstract: Large organizations carry a long tail of infrastructure migrations that are worth doing but rarely urgent enough to compete with feature work, because each one costs significant engineering time per repository. This paper reports on our industrial experience from Automated Uplift, a multi-month initiative applying LLM-assisted migration pipelines across thousands of Bloomberg repositories. We examine two distinct production campaigns: a heterogeneous Python packaging migration (approx. 1,900 candidate repositories) and a uniform C++ build system migration (approx. 6,700 candidate repositories). Together, these campaigns generated 2,333 pull requests, resulting in 729 successful migrations. We detail our hybrid architectural solution which maximizes deterministic processing, confines the LLM to bounded file transformations within a pre-computed context, and incorporates runtime production telemetry (such as execution logs and usage traces). Our findings show that generating migrations turned out to be the easier part; the bottleneck was the cost, for owning teams, of reviewing, testing, and deploying them. These costs were lowered when changes could be verified directly in the pipeline before opening a PR (as with C++ compilation and test validation), and when the population was uniform enough for fixes to carry over from one repository to the next. We conclude by distilling our experience and findings into actionable, transferable patterns for architecting hybrid LLM-deterministic pipelines and scaling code migrations across the enterprise ## Catastrophic Overfitting is Weight Drift the Training Loss Cannot See DevFeed: [Catastrophic Overfitting is Weight Drift the Training Loss Cannot See]() Original publisher: [Read original article]() Author: Chao Pan Published: 2026-11-05T05:00:00Z Content type: article Language: en Sources: [arXiv: Cryptography and Security]() Topics: [Machine learning](), [Adversarial attacks](), [machine learning overfitting]() Tags: [adversarial](), [attack](), [coupling](), [cs-cr](), [cs-lg](), [datasets](), [fast](), [fat](), [moving](), [network](), [overfitting](), [robust](), [robustness](), [training](), [validation](), [weight]() ### AI overview In fast adversarial training, catastrophic overfitting occurs when robustness to multi-step attacks collapses even as the single-step training attack remains repelled. The authors attribute this to weight drift that changes the training objective only marginally, leaving it unable to correct the loss of robustness; clean behavior improves during the same event. They report that the exponential moving average (EMA) weights avoid this drift and remain robust. Their proposed guard resets weights to the EMA when validation robustness drops, alongside a coupling-based base method designed to keep its average within the robust region. In the reported runs, this removes catastrophic overfitting and achieves the strongest AutoAttack accuracy among fast methods on three benchmarks; on CIFAR-100, it also exceeds the multi-step reference at roughly half its cost. ### Source excerpt arXiv:2611.06082v1 Announce Type: cross Abstract: Fast adversarial training (FAT) makes robust training affordable by replacing the multi-step attack with a single-step one, but it suffers from catastrophic overfitting (CO): partway through training, robustness to multi-step attacks collapses to near zero while the single-step attack used for training keeps being repelled. We show that CO is a quiet escape of the weights. Across the event the network's clean behavior quietly improves, and moving from the weights just before CO to those just after it changes the training objective only marginally while destroying most of the robustness. The training objective is therefore nearly blind to the displacement it causes, so it exerts no force to undo it. The exponential moving average (EMA) of the weights, which FAT protocols already maintain, does not accumulate this drift and stays robust through the entire event. This observation suggests a nearly free fix: when validation robustness drops, reset the weights to their EMA. We build this guard on top of a coupling-based base method designed so that its average stays inside the robust region. The resulting training removes CO in every one of our runs and attains the strongest AutoAttack accuracy among fast methods on all three benchmark datasets, including a clear margin on CIFAR-100, where it also clears the multi-step reference at roughly half its cost. Code is available at https://github.com/fzjcdt/co-weight-drift. ## FlyHash for LLM Agent Memory: An Empirical Validation and a New Application DevFeed: [FlyHash for LLM Agent Memory: An Empirical Validation and a New Application]() Original publisher: [Read original article]() Author: Mohan Bheemadolu, Chrisogonas O. Odhiambo Published: 2026-11-05T05:00:00Z Content type: news Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Vectors & LLMs](), [Large Language Model]() Tags: [agent-memory](), [comparison](), [cs-ai](), [cs-lg](), [embeddings](), [hash](), [llm](), [memory](), [memory-retrieval](), [validation]() ### AI overview FlyHash preserves neighbor structure better than SimHash under correctly matched bit budgets, robustly across 10 seeds and four datasets spanning two modalities. The study also finds that this quality advantage does not yield general memory or latency savings: at equal bits, FlyHash codes are larger and slower to compare, and HNSW outperforms both hash methods at ordinary scale. A sparse-set comparison recovers much, but not all, of a naive latency gap. FlyHash's advantage also holds with non-circular agent-memory relevance labels, rather than labels derived from embedding neighbors. The authors report correcting a comparison pitfall and provide an implementation and working demo. ### Source excerpt arXiv:2611.05791v1 Announce Type: new Abstract: Large language model (LLM) agents increasingly rely on external memory stores to retain information across conversations, typically via dense vector embeddings searched with approximate nearest-neighbor methods. The fruit fly's olfactory circuit computes a locality-sensitive hash of sensory input using a strikingly different mechanism - sparse random projection followed by winner-take-all competition - and prior work (Dasgupta et al., 2017) showed this scheme, dubbed FlyHash, preserves neighbor structure better per bit than conventional random-hyperplane hashing (SimHash) on several benchmarks. We apply FlyHash, to our knowledge for the first time, to LLM agent memory retrieval, and ask three questions a practitioner actually needs answered: does the quality advantage hold under a correctly matched-bit protocol (it does, robustly, across 10 seeds and four datasets spanning two modalities, with one specific, previously undocumented comparison pitfall identified and corrected along the way); does that advantage translate into real memory and latency savings, including against the dense-vector indexing (HNSW) practitioners default to today (no: at equal bits FlyHash's codes are larger and slower to compare than SimHash's, and HNSW dominates both hash methods at ordinary scale; a sparse-set comparison we introduce recovers most but not all of a naive latency gap, leaving hashing's practical case narrower and more specific than the quality result alone implies); and does the advantage survive genuine, non-circular agent-memory relevance labels rather than the nearest-neighbor-by-embedding proxy this literature otherwise relies on (it does). We report all three answers precisely, including the two that narrow rather than flatter the headline result, and release our implementation, a working agent-memory demo, and a full account of the methodological pitfalls we found and corrected. ## ToxiEval-ZKP: A Structure-Private Verification Framework for Molecular Toxicity Repair Tasks DevFeed: [ToxiEval-ZKP: A Structure-Private Verification Framework for Molecular Toxicity Repair Tasks]() Original publisher: [Read original article]() Author: Fei Lin, Tengchao Zhang, Ziyang Gong, Fei-Yue Wang Published: 2026-11-05T05:00:00Z Content type: article Language: en Sources: [arXiv: Cryptography and Security]() Topics: [Zero-knowledge proof](), [Cryptography]() Tags: [artificial-intelligence](), [circuit](), [classification](), [complete](), [cs-cr](), [evaluation](), [framework](), [generative](), [hashing](), [model](), [privacy](), [private](), [repair](), [science](), [security](), [structure](), [system](), [validation](), [verification](), [zero-knowledge](), [zero-knowledge-proof]() ### AI overview ToxiEval-ZKP lets model developers prove to external verifiers that generated molecules satisfy multidimensional toxicity-repair criteria without disclosing the molecular structures. Its general-purpose circuit supports classification and regression, combining evaluation logic with Poseidon-based commitment hashing and a nullifier to prevent replay. The abstract reports validation under complete structural invisibility and claims circuit efficiency, security, and adaptability; it gives no quantitative results or detailed limitations. ### Source excerpt arXiv:2508.12035v3 Announce Type: replace Abstract: In recent years, generative artificial intelligence (GenAI) has demonstrated remarkable capabilities in high-stakes domains such as molecular science. However, challenges related to the verifiability and structural privacy of its outputs remain largely unresolved. This paper focuses on the task of molecular toxicity repair. It proposes a structure-private verification framework - ToxiEval-ZKP - which, for the first time, introduces zero-knowledge proof (ZKP) mechanisms into the evaluation process of this task. The system enables model developers to demonstrate to external verifiers that the generated molecules meet multidimensional toxicity repair criteria, without revealing the molecular structures themselves. To this end, we design a general-purpose circuit compatible with both classification and regression tasks, incorporating evaluation logic, Poseidon-based commitment hashing, and a nullifier-based replay prevention mechanism to build a complete end-to-end ZK verification system. Experimental results demonstrate that ToxiEval-ZKP facilitates adequate validation under complete structural invisibility, offering strong circuit efficiency, security, and adaptability, thereby opening up a novel paradigm for trustworthy evaluation in generative scientific tasks. ## Caches, Certificates, Streaming Uploads and a Simulator That Types DevFeed: [Caches, Certificates, Streaming Uploads and a Simulator That Types]() Original publisher: [Read original article]() Author: Shai Almog Published: 2026-11-05T00:00:00Z Content type: article Language: en Sources: [CodeName One]() Topics: [backend frameworks](), [Back end](), [keyboard](), [Cache](), [certificates](), [GUI](), [cache-invalidation](), [infrastructure monitoring]() Tags: [backend](), [cache](), [certificates](), [keyboard](), [simulator](), [streaming](), [test](), [uploads](), [validation](), [video]() ### AI overview The excerpt describes Codename One backend additions and simulator features. The backend now includes a bounded, expiring cache with eviction callbacks and metrics; cached OAuth access tokens with pre-expiry refresh; PEM certificate-chain validation against private anchors; centralized exception handling, interceptors and validation; streaming request bodies that spool large uploads to disk; and a proxy restricted to its target and allowed hosts. JSON mapping adds custom codecs and broader type support, while ORM maps selected java.time values to numeric epoch units. Controller test slices start only listed components and use MockMvc without a socket. The simulator can emulate iOS keyboard behavior and Android resize or pan modes, and record video or GIF with captions and touch indicators. Its keyboard behavior is an emulation, so the excerpt recommends a final device check. The backend remains a tech preview. ### Source excerpt Eight backend upgrades, from a bounded cache to certificate validation and test slices, plus a simulator that slides up a virtual keyboard and records video. ## How to Validate PDF Digital Signatures in JavaScript with Timestamps and Long-Term Validation (LTV) DevFeed: [How to Validate PDF Digital Signatures in JavaScript with Timestamps and Long-Term Validation (LTV)]() Original publisher: [Read original article]() Author: Sumathi Uthayakumar Published: 2026-11-04T15:30:00Z Content type: tutorial Language: en Sources: [Syncfusion]() Topics: [pdf](), [certificates](), [JavaScript]() Tags: [digital-signature](), [digital-signature-pdf-javascript](), [digital-signatures](), [document-processing](), [document-processing-javascript-pdf-web-what-s-new](), [file-formats](), [how-to](), [javascript](), [javascript-pdf-library](), [javascript-pdf-signing](), [long-term-validation-pdf](), [net-pdf-library](), [pdf](), [pdf-digital-signature-javascript](), [pdf-signature-validation](), [pdf-signature-validation-javascript](), [pdf-timestamp-signature](), [security](), [sign-pdf-in-javascript](), [syncfusion](), [timestamps](), [validate-pdf-signatures](), [validation](), [web](), [what-s-new]() ### AI overview The Syncfusion JavaScript PDF Library's 2026 Volume 3 release adds signature validation and Long-Term Validation (LTV). The guide describes signing PDFs with a PFX/P12 certificate, adding a trusted timestamp through a TSA callback and asynchronous save, then reloading the signed PDF to embed OCSP or CRL evidence with enableLTV(). Developers can validate one signature or all signature fields, reviewing integrity, certificate-chain trust, timestamps, and revocation results; a single overall Boolean may not capture each signature's status. Long-term verification is not guaranteed: it depends on signature integrity, available certificate and revocation evidence, chain building, and validator trust policies. The text cautions against exposing private keys or certificate passwords in browser code and says applications must implement TSA and revocation network handling. ### Source excerpt Learn how to sign PDF documents in JavaScript, add trusted timestamps, enable Long-Term Validation (LTV), and validate digital signatures using the Syncfusion JavaScript PDF Library. ## Brava Raises $15M to Address the Blind Spots in AI-Powered Cybersecurity DevFeed: [Brava Raises $15M to Address the Blind Spots in AI-Powered Cybersecurity]() Original publisher: [Read original article]() Author: Kevin John Published: 2026-11-04T14:30:03Z Content type: news Language: en Sources: [HackerNoon]() Topics: [Detection engineering](), [Cybersecurity](), [Agentic SOC](), [Security Operations Center](), [Security Information and Event Management (SIEM)](), [observability]() Tags: [ai](), [ai-attack-simulation](), [ai-driven-soc](), [ai-powered](), [attack](), [brava-15m-seed-round](), [brava-security](), [cybersecurity](), [detection](), [integrations](), [multi-agent-security-operation](), [platform](), [raises](), [security](), [siem](), [siem-rule-repair](), [simulation](), [soc](), [soc-detection-validation](), [splunk-and-google-secops](), [telemetry](), [telemetry-gap-analysis](), [validation]() ### AI overview Brava Security raised $15 million to develop its multi-agent AI platform, which tests attack techniques against an organization's security environment and checks whether the resulting telemetry and SIEM detections expose them. It aims to identify missing logs, broken data feeds, and detection gaps, then help rewrite detections and investigate alongside security teams. The platform is designed to work with existing infrastructure and supports Splunk, Google SecOps, CrowdStrike, and more than 70 other sources and destinations; the company says first insights can arrive 10-15 minutes after onboarding. Brava reports Fortune 100 customers and says funding will support product development and North American expansion. These customer, performance, and funding details are attributed to the company in the supplied article. ### Source excerpt Brava Security raises $15M to expand its AI-driven SOC platform for attack simulation, detection validation, telemetry gaps, and SIEM integrations. ## ZDI-26-869: MailEnable Save Directory Traversal Arbitrary File Creation Vulnerability DevFeed: [ZDI-26-869: MailEnable Save Directory Traversal Arbitrary File Creation Vulnerability]() Original publisher: [Read original article]() Published: 2026-11-04T06:00:00Z Content type: news Language: en Sources: [Zero Day Initiative]() Topics: [exploit chaining](), [Exploit](), [file]() Tags: [7-4](), [authentication](), [context](), [cvss](), [directory](), [exploit](), [file](), [files](), [remote](), [results](), [save](), [system](), [traversal](), [validation](), [vulnerability]() ### AI overview Unauthenticated remote attackers can create arbitrary files on affected MailEnable installations because the Save method uses a user-supplied path without proper validation. The advisory says this may be combined with other vulnerabilities to execute code as SYSTEM. MailEnable fixed the issue in version 10.60. The ZDI assigns CVSS 7.4 and identifies CVE-2026-114556. ### Source excerpt This vulnerability allows remote attackers to create arbitrary files on affected installations of MailEnable. Authentication is not required to exploit this vulnerability. The ZDI has assigned a CVSS rating of 7.4. The following CVEs are assigned: CVE-2026-114556. ## Automated Numerical Stability Analysis of Deep Learning Operators DevFeed: [Automated Numerical Stability Analysis of Deep Learning Operators]() Original publisher: [Read original article]() Author: Xinye Chen Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Deep learning](), [Tool]() Tags: [analysis](), [automated](), [cs-ai](), [cs-na](), [deep-learning](), [inference](), [kernels](), [math-na](), [monitoring](), [operators](), [software](), [training](), [validation]() ### AI overview The paper presents a unified tool for stochastic numerical validation of deep learning operators. It applies CESTAC to supported exposed operations and an operator-level data-perturbation approximation to GEMM-like kernels. The authors say it validates operators in one computation pass, identifies sources of numerical instability, and monitors stability during training and inference. They report testing its ability to detect polluted operators with injected instabilities across various tasks. ### Source excerpt arXiv:2607.25494v4 Announce Type: replace-cross Abstract: Finite-precision arithmetic unavoidably introduces numerical approximation errors. Numerical computations may use insufficient precision or an improper formulation, which leads to numerical instability. In this paper, we introduce a unified software tool for stochastic numerical validation of deep-learning operators. The tool follows CESTAC on supported exposed operations and uses an operator-level data-perturbation approximation for GEMM-like kernels. Our developed software not only enables numerical validation with a single computation pass but also detects the sources of numerical instability and provides numerical stability monitoring during deep learning training and inference. We verified its effectiveness on the detection of polluted operators with injected numerical instabilities across various tasks. We believe that our developed method and tools provide valuable insights into developing numerically stable computing kernels, which are particularly critical for numerically stable and efficient deep learning training and inference. ## Development of tools for all-cause mortality prediction in heart failure patients: a systematic review DevFeed: [Development of tools for all-cause mortality prediction in heart failure patients: a systematic review]() Original publisher: [Read original article]() Author: Erik Aerts, Annika Rosengren, Martin Lindgren, Martin Adiels, Yinan Yu, Helen S\\"oland Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Machine learning]() Tags: [cs-ai](), [deep-learning](), [development](), [machine-learning](), [models](), [prediction](), [review](), [standardization](), [stat-ap](), [validation]() ### AI overview The review found that heart failure mortality prediction research increasingly uses machine learning and deep learning, but inconsistent reporting and limited external validation constrain comparisons and clinical evaluation. Of 1,516 publications screened, 56 studies predicting all-cause mortality over 1-60 months met the criteria. In 64% of development cohorts, heart failure was classified as "Unknown/Undefined," and only 23% of studies reported external validation. The authors call for more standardized study design and reporting to support comparison, reproducibility, and assessment of clinical applicability. ### Source excerpt arXiv:2611.03665v1 Announce Type: cross Abstract: Heart failure is a serious medical condition associated with risk of acute deterioration, recurrent hospitalizations and high mortality making accurate mortality prediction potentially valuable for patient care and healthcare planning. Numerous prognostic models have been developed using conventional statistics (CS), machine learning (ML) and deep learning (DL), but variation in study populations, heart failure definitions, prediction horizons and development practices may limit interpretation of applicability. We conducted a systematic review of mortality prediction models published between 2009 to the start of 2026 searching three databases (PubMed, IEEE and Scopus) for studies predicting all-cause mortality over a 1-60 months prediction horizon among patients with heart failure. Data extracted were cohort characteristics, heart failure definitions, model development, cohort size, validation practices, mortality, computational model design and reported performance. The risk of bias and applicability were assessed using the Prediction Model Risk of Bias Assessment Tool (PROBAST). Of 1516 publications identified, 56 met the inclusion criteria. Publication numbers increased over the observation period, alongside increasing use of ML/DL models and decreasing use of CS approaches. Most development cohorts (64%) were classified as Unknown/Undefined heart failure, reflecting inconsistent reporting and terminology that did not correspond with established heart failure guidelines. External validation was reported in 23% of studies. Overall, research in modeling heart failure mortality has grown alongside increasing adoption of ML/DL approaches. Greater standardization of study design and reporting is needed for meaningful comparison, reproducibility and evaluation of models for clinical application. ## ZEMA: confidence-calibrated signed descent with only one state DevFeed: [ZEMA: confidence-calibrated signed descent with only one state]() Original publisher: [Read original article]() Author: Ali Dadsetan, Domenic Rosati, Frank Rudzicz Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [scaling laws]() Tags: [confidence](), [cs-lg](), [fine-tuning](), [gradient](), [machine-learning](), [momentum](), [muon](), [pretraining](), [private](), [state](), [uncertainty](), [update](), [validation](), [variance]() ### AI overview ZEMA adapts signed gradient descent to how strongly a batch supports each coordinate's gradient sign. It computes a per-coordinate z-score from current-batch per-example gradients, converts that score into a sign probability, then smooths the update with one exponential moving average. Forming the probability before smoothing gives ZEMA one state per parameter, compared with Adam's two, but the abstract reports 4.6x higher per-step time. Required per-example variance can be estimated from a small fraction of the batch. In fine-tuning and pretraining experiments, ZEMA is reported as competitive with Adam, with less validation-loss variability and, in some settings, more consistent recovery of the full-dataset gradient's sign. The abstract proposes extending confidence adaptation to methods such as Muon and to differentially private training. ### Source excerpt arXiv:2611.04987v1 Announce Type: new Abstract: Recent work characterizes the Adam optimizer as sign descent modulated by gradient uncertainty: the update is the sign of the batch gradient, scaled by how strongly the batch supports that sign. We apply this view directly and introduce ZEMA (Z-score Evidence Momentum Adaptation), which measures that support with a per-coordinate $z$-score computed from the per-example gradients in the current batch, converts it into a sign probability, and smooths the resulting update with a single exponential moving average. Because the probability is formed before smoothing rather than after, ZEMA keeps one state per parameter instead of Adam's two, at a $4.6$ increase in per-step time; the per-example variance it requires can also be estimated from a small fraction of the batch. In fine-tuning and pretraining experiments, ZEMA is competitive with Adam, reduces the variability of the validation loss, and in some settings recovers the sign of the full-dataset gradient more consistently. We close with two directions this opens: extending confidence adaptation beyond the sign geometry, to methods such as Muon, and to differentially private training. ## Validation-Gated ML Pipeline Synthesis with Expert-Grounded Self-Evolving Skills DevFeed: [Validation-Gated ML Pipeline Synthesis with Expert-Grounded Self-Evolving Skills]() Original publisher: [Read original article]() Author: Zixiao Zhao, Jing Sun, Zhe Hou, Mengze Li, Qian Liu, Cheng-Hao Cai, Zijian Zhang, Xin Wang, Gillian Dobbie, Jin Song Dong Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Software Engineering]() Topics: [Machine learning](), [AI Chat](), [AI Integration Strategies]() Tags: [agentic](), [automl](), [classification](), [code-generation](), [command-line](), [cs-ai](), [cs-se](), [curator](), [datasets](), [denoising](), [error-propagation](), [evolution](), [execution](), [executor](), [framework](), [full](), [machine-learning](), [ml](), [pipeline](), [skills](), [synthesis](), [validation]() ### AI overview EvoML synthesizes machine-learning pipelines through four phases governed by an Orchestrator that advances only after structural, execution, and semantic checks pass, keeping artifacts auditable and preventing errors from propagating between phases. An Executor applies reusable, expert-grounded skills based on the nnU-Net workflow with online adaptation; after each run, a Curator adds recovered scripts and guidance to the skill library. In evaluations across eight datasets and six real-world task types, EvoML reportedly automated all runs and achieved the best final model quality among the tested agentic AutoML systems and command-line coding tools. Reported results include up to 0.94 Dice for segmentation, 0.99 accuracy for classification, 0.95 SSIM for reconstruction, and 0.79 normalized PSNR for denoising. The abstract does not provide further evaluation details. ### Source excerpt arXiv:2611.04180v1 Announce Type: cross Abstract: Recent advances in large language models have enabled agentic systems that automate machine learning pipeline construction from minimal user input. However, existing agentic AutoML methods either depend on rigid templates that limit adaptability, target narrow task domains, or rely on free-form code generation that lacks reliability guarantees, leaving intermediate errors to propagate silently and degrade the final model. We present EvoML, an agentic framework for end-to-end ML pipeline synthesis built on expert-grounded skills that evolve across runs, executed through a validation-gated harness. The framework decomposes pipeline construction into four phases coordinated by three agents: an Orchestrator that enforces validation gates, advancing each phase only when its outputs pass structural, execution, and semantic checks, which precludes error propagation and keeps every artifact auditable; an Executor that runs expert-grounded skills, grounded in the nnU-Net workflow and pairing reusable templates with online adaptation; and a Curator that performs post-run evolution, consolidating the scripts and guidance recovered during a run into the skill library so it improves across datasets. We evaluate EvoML on eight datasets spanning six sophisticated, real-world ML task types, against agentic AutoML systems and command-line coding tools. EvoML achieves full automation and the best final model quality on all eight, spanning segmentation (up to 0.94 Dice), classification (0.99 accuracy), reconstruction (0.95 SSIM), and denoising (0.79 normalized PSNR), while transferring to broader ML tasks. ## LLM Analysis of 150+ Years of German Parliamentary Debates on Migration Reveals a Shift from Post-War Solidarity to Anti-Solidarity in the Last Decade DevFeed: [LLM Analysis of 150+ Years of German Parliamentary Debates on Migration Reveals a Shift from Post-War Solidarity to Anti-Solidarity in the Last Decade]() Original publisher: [Read original article]() Author: Aida Kostikova, Ole P\\"utz, Steffen Eger, Olga Sabelfeld, Benjamin Paassen Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Large Language Model](), [Data analysis](), [Machine learning]() Tags: [2015](), [annotation](), [bias](), [cs-cl](), [cs-cy](), [cs-lg](), [dsl](), [evaluation](), [fine-tuning](), [germany](), [gpt](), [gpt-oss](), [inference](), [llm](), [migration](), [political-discourse](), [reduce](), [trends](), [validation]() ### AI overview The study applies theory-guided LLM annotation to German parliamentary debates from 1867 to 2025, evaluating model performance across sizes, prompts, fine-tuning, and historical versus contemporary texts. GPT-5 and gpt-oss-120B achieved macro-F1 scores comparable to human agreement, but systematic errors could distort trend estimates. The authors combine soft labels with Design-based Supervised Learning to reduce that bias. Their analysis finds relatively high solidarity toward migrants in postwar Germany, especially group-based and compassionate forms, and a marked rise in anti-solidarity since 2015. They conclude that LLMs can scale social-scientific text analysis, provided outputs are rigorously validated and statistically corrected when errors warrant it. ### Source excerpt arXiv:2509.07274v5 Announce Type: replace-cross Abstract: Migration has been a core topic in German political debate, from postwar expellee displacement to labor migration and recent refugee movements. Large-scale analysis of such political discourse has traditionally required extensive manual annotation, limiting coverage. Large language models (LLMs) offer a scalable alternative. Using a theory-driven annotation scheme, we examine how well LLMs annotate subtypes of solidarity and anti-solidarity in German parliamentary debates and whether the resulting labels support valid downstream inference. We first evaluate multiple LLMs across model size, prompting strategies, fine-tuning, historical versus contemporary data, and systematic errors. The strongest models, especially GPT-5 and gpt-oss-120B, achieve macro-F1 scores comparable to human agreement, although their systematic errors can bias downstream results. We therefore combine soft-label model outputs with Design-based Supervised Learning (DSL) to reduce bias in long-term trend estimates. Beyond the methodological evaluation, we interpret the resulting annotations from a social-scientific perspective across the Reichstag (1867-1933), West German Bundestag (1949-1990) and Re-Unified German Bundestag (1990-2025) corpus to trace trends in solidarity and anti-solidarity toward migrants, with detailed analysis of postwar Germany (1949-1957) and contemporary Germany (2009-2025). We find relatively high levels of solidarity in the postwar period, especially in group-based and compassionate forms, and a marked rise in anti-solidarity since 2015. We argue that LLMs can support large-scale social-scientific text analysis, but their outputs require rigorous validation and, where systematic errors are present, statistical correction. ## Auditing a Model-Hub Benchmark: Data Quality, Training Budgets, and Side Information in HuggingBench DevFeed: [Auditing a Model-Hub Benchmark: Data Quality, Training Budgets, and Side Information in HuggingBench]() Original publisher: [Read original article]() Author: Michal Podstawski Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Data Quality](), [Machine learning]() Tags: [artifacts](), [auditing](), [benchmark](), [budget](), [cs-lg](), [data](), [data-quality](), [data-validation](), [dataset](), [join](), [knowledge-graphs](), [recommendation](), [replication](), [resource](), [text](), [training](), [validation]() ### AI overview The HuggingBench audit finds that valid files can still link different objects: three of its four external knowledge graphs violate the item-entity identity constraint used by KG recommenders. Correcting those joins improves every KG-based method across all three seeds, by up to 73% in mean Recall@10, and reverses ten of 27 KG-configuration comparisons. The 100-epoch budget also appears too short for seven to eight of 29 validation-selected settings; 21 of 23 affected setting-seed pairs improve with longer training. Released item text, unused by the reported methods, provides the strongest signal: closed-form linear models using text and the item graph exceed the best reported Recall@10 by 18% and fit in minutes on a CPU. Checks of other releases find incompatible identifiers for the same dataset. The authors report releasing corrected files, mappings, validators, and logs. ### Source excerpt arXiv:2611.03727v1 Announce Type: new Abstract: Benchmarks built from model hubs combine interaction data, knowledge graphs (KGs) and resource documentation, but valid files do not ensure that these artifacts describe the same objects. Treating the release as a data product, we audit HuggingBench (SIGIR 2025), a benchmark for recommending Hugging Face resources, through data validation, a replication of its eleven recommendation methods and controlled experiments with three training seeds. In three of its four external KGs, the files pass key and range checks but violate the item-entity identity constraint on which KG recommenders rely. Correcting this join improves every KG-based method in every seed, by up to 73% in mean Recall@10, and consistently reverses ten of 27 comparisons between KG configurations. The protocol matters as well: the 100-epoch budget ends before early stopping in seven to eight of 29 validation-selected settings, and 21 of 23 such setting-seed pairs improve with a longer budget. The released item text, which no reported method uses, is the strongest signal: closed-form linear models with text and the item graph exceed the best reported Recall@10 by 18% while fitting in minutes on a CPU. Checks of other recommendation releases show that individually valid releases of one dataset can use incompatible identifiers. We release corrected files, mappings, validators and logs, and derive practices for preserving identity and provenance across the artifacts of benchmarks and model lakes. [Next page]()