Audio deepfake detectors that work well on clean lab recordings often degrade after telephone or platform compression. This paper asks a specific question: which low-dimensional acoustic descriptors stay informative after codec degradation? We screen 36 utterance-level features(formants, PNCCs, LFCCs, prosody) on ASVspoof 2019 data and test on the official ASVspoof 2021 LA set across seven codec conditions. Of 33 statistically significant descriptors, only four show large practical separation (|r| >0.5, bootstrap 95% CIs), all first-formant (F1) measures, highly redundant (PC1: 79% variance). The payoff is stability: the 36-feature SVM degrades 14 pp under codec mismatch (9.5%-23.5% EER), while the four-feature RBF-SVM degrades only 5.9 pp (17.0%-22.9%). The F1 subset varies 3.4 pp across conditions versus 13.1 pp for the full set. F1 differences persist after controlling for speaker identity (p greater than 10^10, mixed-effects model). The results suggest that physically interpretable F1 statistics offer a compact, codec-stable diagnostic signal that may complement higher accuracy neural detectors.
When an LLM agent fails part-way through a long task, the usual analysis records a symptom: a failure-mode label, or the agent and step judged responsible. It does not say where the error entered, which state carried it to the symptom, whether anything in the trajectory contradicted it, or which control could have caught it. We define an episode record that ties a candidate fault to its origin, propagation, detection, response and consequence, located by operational phase and state domain, and that names the control families whose observation and reach cover it. We turned the record into a written annotation protocol and applied it to 95 failed runs of six multi-agent systems from the public Who&When and MAD corpora, reading each trajectory in full. Faults enter early (median at 0.18 of the run), mostly as wrong decisions or fabrications over task state, and in 37 of 88 Who&When runs before the step the corpus' human labels mark as decisive. In 72% of failed runs the trajectory already contained evidence against the error before the failure; in 88% of those no agent acted on it. Something flagged a deviation in 31% of runs, but no response removed the deviated state. Independent judges reading the unused evidence, task contracts fixed at admission, and re-checking beliefs against the environment reach most episodes; restoring the agent's own state would have sufficed in 35 of 95 runs. A text scan of the raw corpus also shows that 45% of the hand-crafted Who&When failures ended with a provider content-filter crash after which the harness emitted a final answer no agent produced, a failure class that neither symptom taxonomies nor responsible-agent labels represent. Protocol, records and scripts are included.
A person running a bench experiment works from a printed instruction sheet, tells identical containers apart by the labels written on them, and looks at the bench again before each step. Lab automation usually works the other way around: the procedure is written as a script, and containers stand at fixed positions or carry fiducial markers. We present Curie's Jarvis, a system that lets a robot arm carry out a chemistry experiment on a bench set up for people. It reads the steps from a photo of the printed sheet, finds containers on an ordinary bench with no fixed positions or markers, tells identical-looking containers apart by reading their labels in four camera views, locates its target again before every step, and sets how each step is carried out, including its speed, from what is being handled. On a Franka Panda arm, Curie's Jarvis completed 5 of 5 runs of one protocol from a children's chemistry kit and 4 of 5 runs of another, with no human intervention, and in 160 locates on printed and handwritten labels it identified every container correctly. This brings robots a step closer to doing bench chemistry the way people do.
Multi-teacher on-policy distillation assigns each sample a domain teacher but typically updates shared student parameters. We introduce Domain-Routed Delta Optimization (DRD-MOPD), which specifies which parameter components each domain's samples update. The student is the sum of a frozen base and trainable components initialized from expert-minus-base weights. All components participate in every forward pass; four update modes define the correspondence between sample domains and components, each with independent optimizer states. We analyze these update rules and measure retention of the original expert geometry through angles, projection coefficients, and energy. Prior work such as Open-MOPD distills directly from the base model; 97.02% of its released final checkpoint's increment energy lies outside the span of the original expert directions. In comparison, our Own checkpoint after 20 updates retains coefficients near one along each component's original direction and pairwise angles ranging from 88.04 to 92.41 degrees; its exported BF16 student's increment has 4.90% outside energy, versus 10.52% for the sum of saved FP32 components. The same checkpoint achieves a 32.72% macro score across six benchmarks, compared with 32.44% for its initial task-arithmetic merge. These measurements describe geometry and capability at the evaluated training stages and settings.
ExpertCrowd-SWE (EC-SWE) is a benchmark of 100 expert-authored, multi-file software engineering tasks for evaluating autonomous coding agents in large, unfamiliar codebases. Each task is a change request written by a practising engineer against a pinned commit of a real repository, paired with a sealed reference solution and a task-specific grader whose weighted, mostly executable criteria award partial credit; must-pass blockers and pass/fail gates encode what a maintainer would refuse to merge. The tasks span 87 open-source projects and 11 languages. The median ticket is 148 words, while the median reference solution changes 471 lines. Every task image is calibrated before release so that the reference solution scores 1.0 and both the untouched repository and an inert change score 0; nearly half of the tasks also ship alternative, lazy and adversarial control patches that the grader must classify correctly. The final 100 tasks were selected for difficulty against current frontier models. We evaluate seven frontier models, five runs each (3,495 scored runs), in a single platform harness that gives each model its provider-native tool interface. Claude Opus 5.5 (67.3%) and GPT-6 Astra (63.7%) lead without a statistically significant gap, followed by Claude Fable 5.1 (46.5%), Gemini 3.8 Flash (40.0%), and a statistically indistinguishable group of Qwen 3.8 Max, DeepSeek V4.1 Flash and Grok 4.6 (32-34%). No task is fully solved by every model, and on 80 tasks no model earns full credit on all five runs. Scores span 35 points. Depending on the model, agents take a median of 33 to 166 steps per run, with little relation to score. We release 24 tasks with their environments, graders and results, and hold out 76 to limit contamination.
Automated video capsule endoscopy (VCE) classification remains challenging due to severe class imbalance, limited source-video diversity, and variations in image appearance across datasets. Although deep learning models have demonstrated promising performance on established benchmarks, their reliability under different validation protocols and external dataset shift remains insufficiently understood. In this work, we introduce DMSNet, a deformable adaptive multi-scale edge-guided attention network integrating adaptive multi-scale feature fusion, deformable lesion refinement, edge-guided attention, and sparse feature selection. We benchmark DMSNet against 18 representative convolutional, transformer-based, and pretrained vision architectures on Kvasir-Capsule under random frame-level and source-video-disjoint validation protocols. We further identify source-video overlap between the original training and test splits and evaluate model performance on both the complete test set and a source-video-disjoint subset. Under random frame-level validation, DMSNet achieves an MCC of 0.442 $\pm$ 0.020 on complete test set and 0.456 $\pm$ 0.037 on the source-video-disjoint subset. On the external CV2024 dataset, ConvNeXt-Tiny achieves the highest MCC of 0.207, while performance degrades substantially across architectures relative to in-domain evaluation. We further investigate source-video diversity using a per-class cap of 200 training frames. With the total training-set size held fixed, increasing the maximum source-video coverage from four to eight improves MCC for 17 of 19 evaluated architectures. Overall, these findings demonstrate that model rankings and classification performance depend substantially on evaluation protocol, training-source diversity, and cross-dataset shift, emphasizing the need for source-aware validation and rigorous external evaluation of VCE classification systems.
We present GigaEdge Spench, an open-source SDK for streaming speech enhancement on CPU servers, mobile devices, and in browsers. Applications can build enhancement channels from neural denoising, normalization and dynamics blocks, sample-rate conversion, and acoustic echo cancellation where supported. These stages can be combined, while normalization can also run without a neural model. The release includes an 8 kHz INT8 denoiser and discriminative and hybrid 48 kHz FP32 models. A shared C++ core supports ONNX Runtime and TensorFlow Lite through native interfaces, Android and iOS SDKs, and WebAssembly integration. Independent stream contexts share model resources, and consecutive frames from one stream can be processed in a single inference call. We evaluate the released models on EARS-WHAM v2 and compare their quality and processing costs with RNNoise and DPDFNet2. All tested native and Web configurations process files faster than real time, although larger chunks require more time to collect input. We also measure how many streams a shared model can process on a CPU. Our contribution is a configurable deployment system and an experimental study of how model choice, chunk size, and execution environment affect its use. All reported SDK timing uses ONNX Runtime. Index Terms: speech enhancement, streaming inference, cross-platform deployment, temporal frame grouping, computational efficiency
In remote inference, edge devices transmit high-resolution images over wireless links to a receiver, where an artificial intelligence (AI) model answers task-specific questions about these images to enable reliable decisions. These answers are often determined by a small set of task-causal pixels, such as those of text, whose corruption changes the answer, so these pixels must be delivered with high fidelity. Conventional transmission protects every pixel equally, so it either consumes many resource elements (REs) or corrupts these regions at low signal-to-noise ratio (SNR). To overcome these limitations, in this paper, a zero-shot, domain-agnostic framework, termed semantic-criticality tiered allocation of resources (STAR), is proposed. In particular, a per-pixel semantic-criticality indicator (SCI) is derived from the attention of a frozen vision-language model, so new tasks and domains are handled by changing only the prompt, without retraining. The resulting task causality-aware resource allocation problem over protection tiers is solved exactly by a threshold rule on the SCI, realized by a coded tier for the task-causal pixels and an uncoded, downsampled base layer for the full image. Simulations over a 5G New Radio link demonstrate that STAR attains a semantic accuracy of at least 0.98 for $E_s/N_0 \geq -4$ dB and, at comparable semantic reliability, reduces the REs of robust coded transmission by 59.5% and 91.5% on factory- and vehicle-monitoring datasets, respectively, the latter without any change to the models. Although a deep joint source-channel coding baseline uses fewer REs at comparable semantic accuracy, it must be retrained with labeled data for every new task or domain.
We present the Neural Control Plane (NCP), a closed-loop system that observes language-model execution, synthesizes candidate weight-space intervention rules, evaluates them, and maintains a versioned table of admitted policies. NCP separates a pretrained model's inference data plane from the control policy that selects temporary parameter interventions. W-ISA supplies transactional execution and parameter restoration; NCP supplies telemetry, candidate synthesis, admission criteria, priorities, and rule lifetime management. The optimization state consists of active rules and an archive of evaluated candidates, while base-model parameters are restored after each request. In a local SmolLM-135M-Instruct experiment, NCP generated three candidates from development telemetry and evaluated them on separate task examples and a 4,095-target WikiText-2 slice. One projection reduced validation token repetition from 0.4133 to 0.2191, with a perplexity change of -0.0574%. All candidates produced zero newly correct exact answers, so the configured admission gate archived all three and retained the original active policy. Fresh-process repetition and saved-table resumption reproduced the measured outcomes. NCP provides an executable framework for studying recursive self-improvement at the control-policy level through explicit synthesis, verification, and state transitions.
Large language models (LLMs) are increasingly studied for clinical decision support, and medical question answering provides a testbed for evaluating this capability. In particular, medical multiple-choice questions (MCQs) reflect a core challenge in clinical decision-making: plausible options often share overlapping evidence, while only a few decisive factors distinguish the correct option from incorrect ones. However, existing retrieval-based and reasoning-oriented methods mainly focus on supplying broadly relevant knowledge or eliciting reasoning paths, rather than identifying the compact differential evidence that separates confusing alternatives. In this work, we present a retrieval-grounded differential evidence framework for medical MCQs. Our method extracts option-specific pieces of differential evidence and organizes them into concise evidence prompts for answer selection. This design better aligns external knowledge with differential reasoning while reducing irrelevant context. Experiments show consistent accuracy gains across multiple datasets and target LLMs. Ablation studies further show that our framework outperforms both directly using retrieved evidence chunks and closed-source LLM-generated general hints, and that its improvements are not solely explained by answer-proximal guidance, but are consistent with differential evidence serving as a compact aid for reasoning over competing alternatives. Our findings also offer insights for medical reasoning and other decision-making settings with multiple possible choices.
We present W-ISA, a software instruction set and transactional runtime for executing weight-space intervention programs on pretrained language models. Shape-checked instructions express parameter updates, projections, ablations, and rank-one transformations. The runtime applies these programs within request-scoped execution contexts and restores saved parameters after completion or interruption. Activation-based match-action rules connect measured model states to intervention selection. We evaluate restoration and replay on SmolLM-135M-Instruct and Qwen2.5-1.5B-Instruct, and compare materialized rank-one updates with equivalent PyTorch hooks on the latter. In 45 timed requests, both intervention implementations produce identical greedy sequences at 32, 128, and 512 tokens; their maximum prompt-logit difference is 0.000184. On the complete WikiText-2 raw test split, the combined SmolLM attention-head ablation and MLP projection changes perplexity from 22.2353 to 22.3188, a relative increase of 0.3754%. W-ISA separates intervention specification, execution, and measurement, making parameter-level experiments reproducible and composable.
In this work, we develop an efficient Gauss-Newton optimization approach for finite basis physics-informed neural networks (FBPINNs). The key ingredient is the overlapping domain decomposition architecture of FBPINNs, which localizes the influence of each subnetwork and induces block sparsity in both the residual Jacobian and the corresponding Gauss-Newton matrix. We characterize this structure and exploit it through sparse automatic differentiation, using graph coloring to partition structurally orthogonal Jacobian columns or rows. This reduces differentiation costs and memory requirements while preserving all structurally nonzero derivative information. The resulting compressed Jacobian representations allow us to assemble the Gauss-Newton system directly in block-sparse form, enabling the use of sparse direct solvers and preconditioned iterative methods. In this way, the otherwise expensive second-order optimization problem is reduced to scalable sparse linear algebra. We demonstrate the approach on multiscale PDEs, where accurate approximation requires rich neural network parametrizations and large numbers of collocation points. Numerical experiments show that our block-sparse Gauss-Newton method attains high accuracy and substantially accelerates optimization relative to first-order baselines while remaining computationally tractable for large problem sizes.
Multi-horizon forecasting has been dominated by discrete forecast strategies that treat model outputs as fixed-vectors or recurrently create them. In this paper, we introduce the Continuous Horizon (CH) strategy that generalizes upon recursive, direct, and multi-step strategies by redefining forecasts as a continuous operator and the forecast horizon as a query variable rather than an index. We present the theoretical foundations for the CH strategy, a Universal Operator Approximation Theorem, and a continuous-horizon bias-variance decomposition that formally proves the estimation variance improvements of CH over the discrete strategies. We validate the CH strategy empirically through a simulation study, and on 16 large-scale datasets from the M-series benchmark, showing how CH achieves competitive RMSE accuracy, while uniquely enabling parallel inference, targeted horizon querying, and interpolation and extrapolation across unseen horizons.
Reasoning models write a reasoning trace at every turn of a multi-turn tool session, and later turns read this history. Across 94k ShareGPT conversations, 73 percent of user interactions span multiple turns. Upstream providers bind each trace to its issuer, so a request forwarded across issuers arrives with its history deleted, and the task cost of this stripping remains unmeasured. To the best of our knowledge, this paper presents the first direct, blind, multi-seed measurement of full reasoning-history deletion in multi-turn tool calling, through three rendering conditions, seven open-weight reasoning models, three seeds, 794 tasks, and the official checker. The same deletion spans -4.4 to +51.3 points with both ends excluding zero, and for the one model re-dosed from 0 to 100 percent deletion, the reading varies by at most 1.5 points. Deletion sensitivity is a model-level property. The reading follows the model itself and stays constant across deployments, and a retrained or re-released model is a new model that a two-cell screen, one retention run and one deletion run, re-reads for the cost of two runs. Deletion sensitivity is thus a measurable property of the model, and the reading must enter routing and aggregation as prior information rather than persist as an invisible default cost.
Causal mediation analysis decomposes a treatment effect into indirect pathways through mediators and direct pathways through other processes. Tabular foundation models perform classification and regression through in-context learning, providing estimates of conditional means and propensity scores used in causal mediation analysis. Constructing confidence intervals from these estimates requires accounting for uncertainty introduced by fitting and combining several nuisance functions. We introduce MediPFN, an amortized inference framework that adapts a pretrained tabular foundation model to nuisance prediction and combines its predictions through a cross-fitted influence-function estimator. A separate head learns quantiles of the estimator's standardized error from simulated tables with known indirect effects. Supervised fine-tuning, symmetrization, and independent calibration then produce a confidence interval. We establish coverage bounds for the resulting intervals under standard assumptions. Across experiments, MediPFN attains coverage closer to the nominal level than several comparators. It also reduces the computational cost of uncertainty quantification, running 59.1 times faster than bootstrap in matched comparisons.
Two accounts of how reinforcement learning with verifiable rewards (RLVR) interacts with an imperfect verifier coexist: in one, reward hacking is inherited from the base model; in the other, RL amplifies it. We study both in a controlled environment small enough to compute the entire response distribution exactly. We vary the universality of a leak, or the fraction of prompts that accept the same wrong response, and the verifier's static false-positive rate. At its KL-regularized optimum, RLVR acts as a filter that reweights the accepted set without altering its composition, so the hack rate is inherited exactly. Taking that as a null model, a known decomposition writes the drift from it as a sum over prompts of covariances between reward and the empirical tangent kernel. The terms from other prompts vanish without parameter sharing. With sharing, every prompt that accepts a wrong answer pushes it up and every prompt that rejects it pushes it down through the baseline, in proportion to the model's probability of producing it there. A shared-parameter policy thus drives a leak that few prompts accept below the base rate and amplifies one that most prompts accept. At every convergent learning rate under Adam, SGD, and sampled GRPO, hacking ends higher at high universality than at low, but whether either end crosses the base rate depends on the step size. Independent per-prompt copies remove the dependence. The direction of this push depends on neither the verifier's error rate nor universality alone, but on how much of the model's kernel-weighted probability of a wrong answer falls on prompts that accept it, which frozen base model controls support. Verifier fixes should target the mistakes the model repeats across prompts: two patches that remove equally many errors differ by up to 38 points of true accuracy. We propose this alignment as a hypothesis for why earlier studies disagree.
This study examines the reliability of Claude Code, an agentic deployment of the Claude Opus 4.6 large language model, as a grading system for handwritten university exams. Four research questions addressed agreement with expert human grading on total scores (RQ1), agreement across five question types from multiple choice to evaluative judgement (RQ2), consistency across two independent grading sessions one week apart (RQ3), and the accuracy of AI-generated feedback (RQ4). Claude Code graded the 20-mark mid-term papers of 60 consenting students in a Saudi computer graphics course in a single agentic session, guided by three rubric files. Reliability was estimated on disjoint samples: 20 papers for RQ1, RQ2 and RQ4, and 15 papers for RQ3. Analyses used Pearson correlation, mean absolute error, Bland-Altman analysis, weighted Cohen kappa, intraclass correlation, and a four-category coding scheme separating diagnostic from confirmation feedback. The author, as the sole subject-matter expert for the course, provided the human benchmark, so the design assesses agreement with a domain-expert reference rather than inter-rater reliability. AI grading agreed strongly with human grading on totals (r = 0.97; MAE = 0.85 marks) with a small but significant AI-positive bias of 0.68 marks, showed excellent test-retest consistency (ICC = 0.96), and produced accurate feedback (92.8% diagnostic; 97.8% confirmation). Drawing items were the outlier on three of four measures. Three failure modes recurred: keyword matching without conceptual depth, selection-state ambiguity on corrected multiple-choice answers, and region-of-attention drift on hand-drawn diagrams. The findings suggest that AI grading at this capability level can assist instructors on objective and short-form items under structural safeguards but is not yet ready to replace instructor judgement at boundary cases or on drawing items.
We investigate how Latin American Studies (LAS) has evolved since 1982 in three leading LAS journals. Using network analysis, text mining, and bibliometrics, we trace transformations in disciplinary positioning, authorship, and citation patterns. Our main claim is that LAS has undergone an assimilationist consolidation. The field is now institutionally recognisable, with core LAS journals occupying central positions in disciplinary networks and acting as bridges across fields. Yet this consolidation has reinforced hierarchies privileging Anglophone academic norms. Journals based in Latin America, once central, have been marginalised. The share of Latin American authors has grown, but their profiles converge towards Northern standards.
We study AI-assisted measurement when one text label applies to many administrative records, the target is a monetary total, and the human readings that anchor the machines disagree. The testbed is 121,070 grant-purpose texts from US tax filings (799,675 records, \$126.4B), labelled by a rule-based classifier and three language models, with a designed sample of 651 texts read independently by three people under one codebook. We state exact conditions for a fixed-predictor difference estimator and for a cross-fitted predictor whose folds are assigned to the whole frame before sampling, under Poisson sampling, and assess the implemented fixed-size procedures by simulation: 8 predictors whose own totals span \$11.2-15.0B agree to within \$0.4B once rectified. Two rules that accept machine labels at a stated dollar-weighted error, weighted conformal p-values under a dollar measure and Learn-then-Test on the design estimate of the error, are evaluated on nested held-out folds and on re-draws of the design with synthetic truths; their error control in deployment is not established, and the count-based rule accepts no text under a dollar-drawn sample. In a replay that charges every human label, a two-phase review policy whose calibrator is fitted on its own first phase reduces the RMSE of the total by 28% at a budget of 100 texts relative to dollar-proportional sampling, while pure Neyman allocation fails with an uncalibrated predictor. Independent readers shift the estimated total by \$3.93B, 2.2 times the within-reader sampling interval, and institution-level intervals under-cover at every size. Every headline result is re-run under each reader. The results identify both the benefits and the limits of using learned predictions to decide what a person should verify.
Children's earliest years are a period of intense social learning, during which meaning is built through gaze, gesture, and physical contact. Yet, this critical aspect of social interaction remains under-evaluated in multimodal video understanding, as existing benchmarks prioritize the conversation-driven interactions of adults. We address this gap with \bench ($\textbf{N}$aturalistic $\textbf{E}$arly-childhood $\textbf{S}$ocial in$\textbf{T}$eractions), the first child-centric benchmark for nonverbal social reasoning. \bench comprises \totalvids manually annotated clips ($\sim$12s) curated for diversity across naturalistic settings, cultural contexts, and developmental stages. Beyond a lack of data, targeted model improvement in this domain is hindered by coarse evaluation schemas that conflate basic perception with higher-order reasoning. To resolve this, \bench introduces a compositional schema that builds atomic interaction cues into progressive levels of social reasoning: contextual, behavioral, interpersonal, and open-ended descriptions. Extensive analysis of state-of-the-art Vision-Language Models (VLMs) reveals a stark performance collapse: despite strong contextual recognition, models exhibit substantial gaps in decoding precise behavioral and interpersonal cues. We trace these failures to specific biases, including poor consequence grounding and temporal sensitivity. Finally, we ground \bench in real-world applications by evaluating VLMs using domain adaptation strategies. Ultimately, \bench provides a rigorous testbed for advancing socially grounded artificial intelligence in early childhood. Project page: https://laubravo.github.io/nest_website/.