Papers

Although Large Language Models (LLMs) encode rich factual knowledge in their parameters, reliably recalling and verifying such knowledge remains a key bottleneck in factual question answering. Existing end-to-end methods entangle knowledge elicitation with reasoning, making it difficult to determine whether correct answers arise from parametric knowledge or the input context. To address this challenge, we propose VAKE (Verifiable Activation of Parametric KnowledgE), a two-stage reinforcement-learning framework that externalizes latent parametric knowledge through explicit Priming and transfers the acquired elicitation capability to implicit Reasoning. Given a query and an insufficient retrieved subgraph, the Priming policy explicitly inserts bridging triples as verifiable evidence, with supervision provided by rewards derived from answers generated by a separate frozen model over the augmented subgraph. Building on the policy learned during Priming, the Reasoning stage trains the model to answer from the original input, testing whether the capability acquired through explicit knowledge elicitation transfers to implicit reasoning. Experiments across seven benchmarks and models from 3B to 14B show that VAKE consistently outperforms standard baselines, including when transferring directly from HotpotQA to OOD datasets. LLM-based evaluation further shows that over 80% of the inserted triples provide factual bridging knowledge not derivable from the retrieved context, while more than half elicit knowledge inaccessible through direct prompting. These results suggest that VAKE activates latent parametric knowledge rather than copying the input context or memorizing dataset-specific associations.

Training terminal agents requires scalable executable supervision, yet synthesizing high-quality terminal tasks remains challenging. Each task couples an instruction, an initialized environment, a reference solution, and an executable verifier; if these artifacts are generated from inconsistent assumptions, the resulting task may be unsolvable or incorrectly evaluated. Meanwhile, multi-stage synthesis can discard the goals, dependencies, state transitions, and procedural constraints encoded in the original sources. We present FACET (Fine-grained Agentic Construction of Executable Tasks), a framework that addresses both information preservation and cross-artifact consistency. FACET reconstructs related agent skills into coherent, information-rich scenarios, then realizes and repairs the execution environment before generating the final task artifacts. The resulting container state serves as shared grounding for the instruction, solution, and verifier, while execution-based validation and targeted repair correct artifact-specific failures without unnecessarily regenerating valid components. FACET produces complex terminal tasks with dense executable checks, and successful trajectories collected from these tasks provide effective, data-efficient supervision. Fine-tuning models across multiple scales consistently improves performance on Terminal-Bench 2.1, while analyses of alternative generation schemes support the importance of environment-grounded construction for task validity and solution-verifier alignment. These results establish source-intent preservation and shared executable-state grounding as key principles for scalable terminal-task synthesis.

Multimodal large language models (MLLMs) have shown strong potential for image quality assessment (IQA) by improving consistency between quality ratings and their underlying reasoning. However, most approaches supervise reasoning through human-provided ratings and rarely examine whether it faithfully reflects image quality. Rating accuracy alone does not ensure faithful reasoning; a shared reward also obscures supervision sources and may reinforce unfaithful reasoning when a correct rating occurs by chance. To improve the faithfulness and reliability of blind IQA, we aim to (1) decouple credit assignment for reasoning and rating and (2) provide verifiable supervision for faithful reasoning. We introduce MR-IQA-2, an actor-editor-judge framework that operationalizes reasoning-editing-reflection. The actor generates quality reasoning for an input image, and the editor revises the image according to the identified quality factors. A frozen judge compares the original and edited images and provides reflective supervision for the actor's reasoning. MR-IQA-2 further uses fine-grained credit assignment to decouple reasoning and rating supervision. Judge feedback supervises reasoning, whereas human ratings supervise the predicted rating. Masked token-specific updates distinguish these signals while preserving the causal relation from reasoning to rating. Across IQA benchmarks, MR-IQA-2 achieves competitive rating alignment with humans. Visual reflection also enables richer and more faithful visual understanding beyond rating, which may inform image-quality optimization and related downstream tasks. Code is available at https://github.com/RobinY99/MR-IQA-2.

Spatial aliasing occurs when two or more distinct locations produce highly similar place-cell representations, primarily due to environmental symmetry or repetitive structures. This issue is most pronounced when place representations are constructed solely from boundary vector cell (BVC) inputs, because symmetric or repetitive structures can yield indistinguishable sensory patterns across multiple locations in an environment. This work introduces grid cell signals to mitigate spatial aliasing in such settings. Because grid cells contribute periodic, internally generated spatial signals that vary independently of environmental geometry, they play a key role in disambiguating perceptually identical locations. We integrate multiple modules of analytically constructed grid cells with BVC-driven place cells and show that this leads to a 94–99% reduction in spatial aliasing relative to a BVC-only baseline across three environments: an open environment without obstacles; an environment with a cross-shaped central obstacle creating high visual symmetry; and a maze environment. The greatest improvement occurs in the environment with the highest visual symmetry. These results indicate that grid cells provide information complementary to boundary-based inputs, yielding more reliable place representations in geometrically ambiguous environments.

The prediction of equilibrium beach profiles under tidal influence is of fundamental importance for sustainable coastal development, informing shoreline protection strategies and managing coastal ecosystems under changing environmental conditions. However, it remains challenging due to the highly nonlinear interactions among wave, tide, and sedimentary processes. Traditional empirical and numerical models often exhibit limited adaptability across diverse coastal environments, with especially pronounced limitations in beach systems where tidal processes are important . To improve data-driven prediction under these conditions, this study proposes MorphoGP, a unified category-specific Gaussian process framework for predicting equilibrium beach profiles (EBPs) under tidal influence. The framework first introduces a ContourCluster model based on contrastive learning to classify tide-influenced beach morphologies automatically. Within each morphological category, a specialized Gaussian process expert learns statistical associations between environmental descriptors including waves, tides, and sediments and the beach profile's shape. A Gating Net then integrates the outputs of all experts through a probabilistic weighting mechanism to produce the final prediction. Evaluated on data from over 180 beach profiles from tide-influenced coasts along the Chinese coast, MorphoGP achieves improved predictive performance compared with conventional and deep learning models, reducing the test RMSE by about 59.3% compared with the best baseline and achieving a final RMSE of 0.297 m. The proposed framework provides a physically informed, data-driven tool for equilibrium beach-profile prediction under tidal influence and coastal management, while stronger process-level physical coupling remains an important direction for future development.

Machine Learning as a Service (MLaaS) is a powerful cloud paradigm enabling data-driven intelligent applications in Internet of Things (IoT) environments, widely adopted across healthcare, smart homes, and industry due to its cost-effectiveness. However, the dynamic nature of IoT frequently alters data distributions, affecting MLaaS stability, while periodic MLaaS updates further introduce performance drift. Unlike traditional ML systems, MLaaS clients operate as black-box users without access to internal data or parameters, making drift detection particularly challenging. To address this, we propose a novel MLaaS Performance Drift Detection framework for IoT environments. The framework first employs an MLaaS extraction model that learns service behavior from input-output pairs and identifies prediction-influenced features. Building on this, the proposed MLaaS Performance Drift Detection (MPDD) model jointly captures variations in input data and MLaaS behavior. We further design an Adaptive-Temporal Performance Drift Detection Mechanism (APDDM) that dynamically adjusts monitoring frequency based on behavioral and data variations, enabling timely drift detection for effective service management. Extensive experiments on real-world datasets demonstrate that MPDD achieves up to 22-25% accuracy improvement over baseline drift detection methods. APDDM provides an average accuracy gain of approximately 4% and reduces the miss detection rate by around 9% compared to fixed-interval monitoring.

Most LLM benchmarks rank models on their ability to automate work tasks. In practice, however, models are often used to assist other (human or LLM) agents. The question that drives model selection is therefore not only which model produces the best output, but which model most improves the work of another (weaker) agent. We introduce a unified framework that evaluates the capability of models to automate and augment another agent's performance. Across seven economically grounded real-world tasks, an assistant model writes assistance text for a standardized lower-capacity worker model, which produces the deliverable. In automation mode, the assistant produces the output directly. Outputs are scored through blind pairwise comparisons by an LLM judge panel with task-specific rubrics, replicated across ten runs. Rankings across the two regimes are only modestly correlated, and the automation winner loses augmentation on five of seven tasks. Assistance is not reliably positive. The unaided worker outranks every assisted condition on three tasks, and only one model's guidance beats no guidance on average. These results suggest that automation ability is an incomplete proxy for assistance quality, motivating benchmarks that evaluate models according to the roles they play in human-AI and multi-agent systems.

Modern e-commerce platforms often operate search, recommendation, personalization, and CRM systems independently, limiting opportunities for proactive customer re-engagement. This is particularly challenging for exploratory intents such as best smartphones or latest 5G phones, where users may leave the platform for external research before purchasing. We present a scalable, production-deployed framework that bridges search and CRM workflows through AI-powered Product Research Agents. The system identifies users with exploratory purchase intent and low engagement, conducts grounded multi-agent product research using behavioral signals, external knowledge, and enterprise catalog data, and delivers personalized recommendations through WhatsApp. We evaluate the framework in a 23-day production deployment involving approximately 15K WhatsApp notifications for mobile product discovery. The campaign achieved substantial CTR improvements over traditional WhatsApp recommendation campaigns, with evidence of secondary engagement through message forwarding and sharing. The deployment also generated downstream purchases and GMV impact, demonstrating the practical effectiveness of AI Product Research Agents for proactive customer re-engagement and end-to-end customer journey optimization.

The remarkable capabilities of large language models (LLMs) are often undermined by their instability. Even subtle and semantically irrelevant changes in prompts can cause dramatic fluctuations in performance, a phenomenon known as prompt sensitivity. Previous studies typically evaluate prompt sensitivity by comparing the LLM's final outputs when prompts change. However, such coarse-grained metrics fail to explain the internal reasons for prompt sensitivity. In this paper, we introduce interactions as a fine-grained tool to analyze prompt sensitivity of LLMs. Specifically, we decompose the output score of the LLM into a set of interactions. Each interaction represents a nonlinear relationship involving a set of input variables. We discover that subtle changes to prompts can trigger severe instability in interactions, even when the outputs of the LLM remain the same. To this end, we propose an Interaction-based Prompt Sensitivity (IPS) metric by quantifying changes in interactions when we introduce subtle changes to prompts. We apply the IPS metric to 50 open-source LLMs and uncover four factors that reduce the prompt sensitivity of LLMs, including supervised fine-tuning, increased model scales, dense architectures, and few-shot learning. More crucially, we discover a common mechanism by which these four factors reduce prompt sensitivity: all four factors tend to reduce the prompt sensitivity of low-order interactions (i.e., interactions involving few input variables).

Large language models are increasingly used to support financial operations, but their apparent reasoning performance can depend on whether they receive the right evidence. In financial reconciliation, the evidence needed for diagnosis is distributed across invoices, purchase orders, approvals, allocations, payments, ledger entries, and bank activity, linked by transactional relationships rather than textual similarity. End-to-end accuracy can therefore conflate evidence access with reasoning quality. We introduce FinRCA-Bench, a deterministic synthetic benchmark of 2,250 accounts-payable-to-bank reconciliation cases spanning 14 operational tables, including 1,500 injected failures across 15 causal categories and 750 legitimate or hard-negative cases. Root-cause labels and record-level evidence contracts are hidden from the model, allowing retrieval to be evaluated independently of answer correctness. We compare Rules/SQL, classical machine learning, dense semantic retrieval, deterministic relational expansion, and Typed Provenance Graph Retrieval (TPGR), a typed traversal restricted to persisted transaction relationships. Rules/SQL reaches 84.97% held-out exact accuracy and classical ML reaches 95.44%. Holding the reasoning model, prompt, and generation settings fixed while changing only retrieval increases macro required-record recall from 0.83% to 77.70% and exact 16-class accuracy from 2.05% to 72.44%. Structural retrieval failures outnumber reasoning failures with sufficient retrieval by 95 to 15; 254 correct predictions occur despite incomplete retrieval, and strict returned-evidence contract accuracy is only 5.72%. On FinRCA-Bench, retrieval architecture strongly shapes observed AI-system performance, and a correct root-cause label is a weak proxy for an auditable diagnosis.

Industrial explainable-recommendation systems built on LLMs incur a substantial serving cost: each request triggers an LLM generation, with latency in the hundreds of milliseconds and cost that scales linearly with traffic. We separate generation from selection: explanations are produced ahead of time as a frozen candidate pool (six prompt styles, two commodity LLMs), and a small CPU-resident selector picks one at request time. The stack needs no GPU and returns in under 100 ms. Our primary benchmark is a 2,958-pair XRec Google Local subset, evaluating six offline-pool selectors (LambdaRank, PPO, GRPO, DPO, teacher-student distillation) and three KG-path selectors (random walks, edge-disjoint enumeration, MMR-reranked paths). A 300-pair MovieLens-1M split with Claude-Sonnet-4.5 references serves as an internal cross-dataset check, since no public benchmark exists for this setting. All variants use the same BERTScore-F1 protocol as XRec and G-Refer, averaged across five seeds. LambdaRank reaches F1 = 0.500 on Google Local, exceeding both G-Refer and XRec, and F1 = 0.329 on the MovieLens-1M check. With seed variance below 0.003 F1, the ordering is reliable: pairwise learning-to-rank outperforms single-action RL (PPO, GRPO, DPO), which use only one labelled candidate per rollout, leaving K-1 labels unused. The KG-path family targets a different objective: all three variants reach USR = 1.000 on Google Local and 0.997-1.000 on MovieLens-1M, since per-request path grounding yields a unique output per query, avoiding template-collapse failures affecting cached-LLM outputs. A generator-pool study comparing Claude 3 Haiku and Claude Haiku 4.5 shows small F1 shifts (0.001-0.006) while preserving selector ranking: selector and generator can be evaluated independently, though absolute F1 depends on the generator. End-to-end build cost is near $15 on commodity hardware.

Equipping Large Language Models (LLMs) with multi-turn tool-calling capabilities is essential for building autonomous agents. However, progress is fundamentally limited by the reliance on full-length trajectory imitation. For tasks involving multiple order-independent sub-goals, the optimal solution space forms a vast combinatorial diamond lattice. Forcing this rich topology into monolithic trajectories causes a severe topological collapse, indiscriminately penalizing valid alternative explorations and severely degrading policy diversity. To address this, we propose DART-SD (Diamond-topology Aware Retrieval and Tuning for Self-Distillation), a novel framework that shifts the paradigm from global forcing to topology-guided localized correction. DART-SD first models the execution process as a converging Interaction-State Transition Graph (ISTG), faithfully capturing the inherent diamond topology of successful and failed exploratory paths. During autonomous rollouts, the framework identifies the Critical Topological Breakpoint (CTB) and retrieves success-supported recovery references. Finally, we introduce a progressive self-distillation paradigm through CTB-guided localized supervision, ensuring that the training loss is calculated exclusively on the generated recovery steps while strictly protecting the valid reasoning prefix from destructive gradient updates. Experiments on complex multi-turn tool-calling benchmarks demonstrate that DART-SD significantly outperforms traditional full-trajectory baselines.

Diffusion-based generators have made synthetic images ubiquitous, but detectors often fail under simultaneous shifts in generator, prompt/style, and source-domain. We study AI-generated image detection as a transfer system described by training prior, frozen encoder feature space, and decision rule, and ask when classifier head training adds value beyond what is already separable in modern features. As a controlled diagnostic, we fit a prior-conditioned Gaussian discriminant ladder: closed-form heads built from first- and second-order feature statistics under nested covariance assumptions. On Percept-Lens, a unified protocol over 39 public datasets (7.1 million images), the best rung is frequently competitive with, and sometimes exceeds, released AI-generated image detector heads when matched on both prior and encoder. We further quantify strong sensitivity to the training prior, data-efficiency of moment-based heads, and representation dependence of Gaussian shift metrics, motivating (prior, encoder, head)-level reporting and stronger analytical baselines for AIGI transfer.

Large antenna arrays allow wireless systems to serve more users and achieve higher data rates, but they also make channel feedback expensive: the receiving device must repeatedly report a large complex-valued channel matrix to the base station. Most neural compressors treat this matrix like an image and replace it with a fixed-length code that only a matched neural decoder can interpret. The message therefore does not adapt to channel complexity, and changing the antenna count typically requires retraining. We ask whether a device can instead report only the few dominant propagation paths underlying each channel. We introduce the Gramian Chebyshev Neural Operator (GCNO), a physics-based, variable-rate compressor that identifies a sample-dependent set of path directions. GCNO uses receive-transmit channel structure to locate paths, a first-order Taylor correction to refine directions that fall between grid points, and least squares to recover their complex strengths. It is trained without path labels, and the base station reconstructs the channel analytically from the transmitted path tuples rather than through a learned decoder. Across three ray-traced environments, GCNO achieves better reconstruction accuracy at the same payload - or lower payload at the same accuracy - than neural feedback baselines, and transfers to unseen antenna counts without retraining.

Dense-caption retrieval has recently been improved by introducing segmentation, edge maps, LLM-filtered captions, and cross-modal modules into contrastive fine-tuning. However, these methods largely inherit the same InfoNCE objective, whose optimization can prematurely saturate under a strong pre-trained initialization: on dense captions, the loss falls below 10^-3 on 80% of batches within the first epoch, while its gradient becomes numerically zero in 47% of measurements. We find that this behavior is closely related to the large number of near-duplicate captions in dense-caption benchmarks, where a few highly similar negatives remain unresolved after the easy majority has already been separated. As a remedy, we introduce HN-CLIP, which uses the text encoder's own text-text geometry to construct per-negative adaptive similarity margins. Specifically, a detached caption-similarity matrix is added to the negative logits, assigning larger margins to more similar captions without mining, synthesizing, or resampling negatives. The resulting objective requires only one caption-similarity matrix and a masked logit addition during training, with no auxiliary data, additional parameters, offline preprocessing, or inference-time overhead. Extensive experiments on four dense-caption retrieval benchmarks show that HN-CLIP improves over the strongest competitors by +2.5 to +4.0 R@1 while training 2.4x faster than GOAL and 5.4x faster than StructXLIP. Moreover, the proposed objective improves all six tested fine-tuning frameworks on the in-domain benchmarks and reaches the strongest full-data baseline with only 20% of the training data.

Precise segmentation of the optic disc and cup is critical for the early detection and diagnosis of glaucoma. However, achieving consistently high performance across datasets while maintaining low computational requirements remains a significant challenge. In glaucoma detection, low-computation methods are crucial for enabling rapid, large-scale screening and facilitating deployment in resource-limited clinical environments. While deep learning models such as UNets, Vision Transformers (ViTs), and Diffusion models have demonstrated strong segmentation performance but these methods often come with substantial computational overhead. UNets are efficient at capturing local features but are limited in modeling global contextual information. Conversely, ViTs excel at long-range dependency modeling but are computationally intensive. Hybrid architectures, such as UNetR, which combine transformer-based encoders with UNet-style decoders, have shown improved performance but while incurring additional complexity. Considering these, in this work, we propose OptiModNet, a light weight novel hybrid architecture tailored for optic disc and cup segmentation. The model integrates diverse attention mechanisms at multiple stages of the network to enhance both local and global feature representation. We include an Aggregated Pyramid Loss that supervises predictions at multiple decoder depths, to promote better gradient flow and structural consistency. We evaluate OptiModNet on the REFUGE2 dataset for both optic disc and cup segmentation tasks. Our method achieves state-of-the-art performance, exceeding existing approaches by over 2.5%, while maintaining high efficiency with only 3.73 GFLOPs and 1.93M parameters. The code is available at https://github.com/SG1947/OptiModNet.

We are witnessing an explosion of agentic systems for computational chemistry: from four in 2024 to seventeen in 2025 and over sixty now, surveyed here. What is delegated to these systems is shifting from single calculations to whole in silico experiments and even manuscript writing. The ultimate destination is a fully autonomous AI scientist, where the entirety of computational chemistry is performed on a machine by a machine, without human supervision. Our survey shows that these systems are turning into vetted chemistry skills on general-purpose coding agents, and that they must be evaluated not only on their final answers but also on whether their calculations actually support these answers. The role of the human computational chemist is shifting from performing calculations to directing and supervising them, and the field should invest in the judgement that makes the supervision reliable: we propose a reporting standard, an evaluation reproducing published studies, a controlled comparison with general-purpose coding agents, and how to teach.

Universal multimodal retrieval aims to support diverse instruction-aware retrieval tasks, demanding both efficient corpus-scale matching and fine-grained semantic reasoning. Recent MLLM-based embedding methods typically derive representations from hidden states, while Chain-of-Thought (CoT) reasoning is emerging as a promising strategy for embedding enhancement by encoding intermediate semantic evidence into the representation space. However, existing CoT methods typically use item-wise reasoning over queries and candidates in isolation, providing no explicit evidence to distinguish a positive from a semantically confusable hard negative. Moreover, contrastive embeddings capture global similarity but struggle with meta-tasks requiring answer verification, category judgment or fine-grained reasoning. In this paper, we propose UMER, a Unified Multimodal Embedding and Ranking framework for universal multimodal retrieval. UMER replaces item-wise reflection with Pair-Aware Discriminative Reasoning, which compares query–candidate pairs to identify instruction-relevant matching and discrepancy evidence. UMER jointly learns contrastive embeddings for efficient global matching and discriminative ranking for explicit pairwise relevance judgment within a single MLLM. A complementary mutual distillation strategy further transfers reliable pairwise preferences between the embedding and ranking functions. On the MMEB-V2 benchmark, UMER achieves state-of-the-art performance under comparable experimental settings while supporting budget-adjustable inference.

Radio maps are essential for wireless decision-making tasks such as access-point placement, coverage planning, and localization, but their fine spatial details are governed by complex propagation effects and are costly to simulate accurately. Machine learning offers a path to high-fidelity radio-map prediction without running expensive high-fidelity simulations for every scene. However, generating high-quality training labels at scale is also difficult: the affordable labels come from finite-ray simulations, which are richer than low-fidelity inputs but carry residual Monte Carlo noise. We address this challenge with Physics-Unrolled Hybrid Neural Operator (PU-HNO), a three-stage cascade that predicts high-fidelity indoor radio maps from low-fidelity ray-tracing outputs and scene priors by progressively capturing reflection, diffraction, and scattering effects, rather than treating radio maps as generic images. We prove that, under conditionally unbiased label noise, the model can learn stable propagation structure and outperform its own training labels. Experiments across diverse floorplans show that PU-HNO outperforms image-to-image baselines, wireless learning models, and monolithic neural operators across both image-quality and wireless deployment metrics.

Block-sparse attention accelerates video transformers by grouping queries and keys and computing attention only between selected groups. However, it faces two challenges: (1) queries sharing a block route may attend to different keys; and (2) retaining most attention mass does not necessarily preserve the attention output. We find that partition choice affects both the overlap among grouped queries' preferred supports and how well an affine function of the sparse output can represent the difference between dense and sparse outputs. We introduce SparsePR, a training-free method combining Response-Coupled Partitioning with Probe-Fitted Residual Reconstruction. We first group paired key/value tokens by how their keys respond to sampled queries. We then group queries by their responses to the resulting key-group centroids for shared sparse routing. We compute exact attention on a small set of query rows and fit an affine correction to the differences between their dense and sparse outputs using weighted ridge regression. The fitted map predicts a residual for each unprobed query, which we add to its sparse output. We measure attention-output error as the relative L2 error with respect to each query's dense output. Across four video generation and world models, SparsePR reduces this error, averaged over unprobed queries, by 61.6–90.5% compared with semantic block-sparse attention without residual correction, at 22% executed-pair density including exact probes. SparsePR achieves 1.48x–2.61x end-to-end speedups over dense attention, with dense-reference PSNR of 24.42–31.84 dB at 21.9–26.0% realized executed-pair density. Project page: https://pardistaghavi.github.io/SparsePR-website/

Hardware functional verification relies on high-quality assertions to expose design bugs and establish confidence in Register Transfer Level (RTL) designs. Yet existing assertion mining methods still struggle to produce complete and reliable assertion sets: random or limited traces fail to cover hard-to-reach behaviors, and one-shot generation provides little feedback about what remains unverified or how the assertion set should be improved. As a result, critical design behaviors can remain uncovered even when many assertions are generated. We present NeuroAssertion, a coverage-driven assertion generation framework that combines formal trace generation, syntax-guided synthesis (SyGuS), and an agent-inspired refinement process within a unified framework. Our framework first converts hard-to-reach control-flow conditions into formal reachability objectives, uses model checking to generate behaviorally diverse traces, and mines initial assertions from these traces with SyGuS. It then performs targeted agent-inspired refinement under verification feedback: one LLM first proposes candidate assertions for uncovered regions, and if a candidate fails formal checking, a second LLM generates a repair grammar that guides constrained symbolic synthesis in a neuro-symbolic repair procedure. Experimental results show that this framework delivers around 2X more assertions and about 2X higher mutation coverage than traditional assertion mining methods.

Lightweight proxy models enable rapid experimentation without repeatedly training frontier-scale systems, but their small kernels often leave modern accelerators underutilized. Conventional training compounds this inefficiency by scheduling the forward and backward passes as disjoint phases, so spare capacity in one cannot be filled by work from the other. We reinterpret the detachment mechanism of Forward-Forward (FF) as a scheduling primitive: given a local objective, detaching a block's output removes downstream gradient dependencies, making its backward pass ready when its forward pass finishes. ERASE launches each detached subgraph's backward pass early on a separate CUDA stream, overlapping it with subsequent forward work. Execution trace on a lightweight transformer demonstrates this overlap and its limit: a kernel that saturates the device leaves no capacity for concurrency. On a large-scale click-through-rate model, detaching six dense subarchitectures improves training throughput by up to $9.51\%$ while keeping normalized entropy close to the baseline.

We present the first mechanised formalisation of Romanov's Triplet Logic (TLS) in the Rocq proof assistant. TLS is a combinatorial framework originally motivated by Boolean satisfiability, based on triplet structures and a filter that we call Simple Vertex Intersection (SVI). We formalise the core of TLS, including its translation from 3-CNF, the clearing procedure, and the SVI algorithm. For the well-formed sliding-window fragment, we prove explicit polynomial-time bounds for the filter stages and verify the translation and intersection operations. Our main contribution is a precise correctness boundary: for general formulas, SVI non-emptiness is necessary but not sufficient for satisfiability; for aligned structures, we prove a full bi-implication, extended to systems of structures. We also formalise the grouped-window translation and provide a formal counterexample to its completeness. We introduce VFR (Verified Filter for Romanov's triplet logic), an extracted OCaml prototype that implements a verified decision procedure for the sliding-window fragment and a sound filter for general 3-CNF, with a Python runtime and Docker packaging. Benchmarks corroborate the predicted behaviour, and the complete toolchain is available as a curated Zenodo artifact. The Rocq development comprises over 23,000 lines of code, with 424 proved lemmas and no unproved assumptions.

Modern mental healthcare faces a critical shortage of senior supervisory oversight, leading to a "supervision gap" where novice therapists manage high-stakes risks with delayed professional feedback. This paper proposes a new framework utilizing a fine-tuned Mistral-7B-instruct model as an automated "Supervisor-in-the-Loop" system. By leveraging 106 sessions from the DAIC-WOZ dataset, the model performs a tri-stream analysis: (1) Therapeutic Alliance tracking via semantic adherence, (2) Latent risk prediction using attention-weighted analytics, and (3) Supervisory Triage via a Dynamic Clinical Urgency Index (D-CUI). Our multi-modal VAL (Visual-Acoustic-Linguistic) framework achieves 95% technique identification accuracy [95% CI: 75.1%-99.9%], alliance assessment MAE of 0.105 on a 5-point scale [95% CI: 0.059-0.151], therapeutic fidelity alpha = 0.423, and mean D-CUI of 0.370 [95% CI: 0.322-0.419]. Training converged in 105 steps with 85.2% loss reduction on a single Tesla T4 GPU. The system reduces supervisory triage latency from 72 hours to real time (~10 seconds per session), enabling proactive intervention in high-risk cases. The system addresses the cold-start problem through Bayesian priors and implements timestamp-based modality synchronization for robust multi-modal fusion.

Language model agents now execute bounded tasks reliably. Whether they can sustain effective decision-making over long horizons, where actions have cumulative consequences and the environment responds to their choices, remains largely unmeasured. FM-Bench (Football Management Benchmark) measures this. An LLM agent runs a football club for 20 in-game years through 26 tools and roughly 340 to 400 decision stops. It drafts a squad on the same budget as every rival, trades players, negotiates contracts, invests in facilities and youth, sets lineups, and answers to a board that can fire it, while a deterministic engine accumulates every year into one final score with no LLM judge or human rater. The solo track plays each of 15 frontier models against a frozen scripted world, and the Arena places the same models plus a scripted anchor in one shared 20-year world; to our knowledge, the first head-to-head evaluation at this scale. We measure six behavioral capabilities behind the score. Across three seeds, all 15 models complete every horizon while the blind scripted baselines die out in most of theirs, and claude-fable-5 tops the solo board on mean score and the Arena, where the title nonetheless rotates among ten models. Neither scale, price, nor vendor predicts the order; the order settles only late in the horizon, and the best first-play human lands only at the bottom of the model board. What separates the models is managerial behavior rather than computation. Higher-scoring models reduce slow-payoff investment near the end, keep cash invested rather than idle, and open renewals well before the deadline, while token spend predicts nothing. No model learns the market's hidden prices from hundreds of rejected bids, and self-managed memory fails in two opposite modes: an archive that only grows or a plan rewritten every season. Code is available at https://github.com/Analogy-AI/fm-bench.

Recent work has shown that large language models (LLMs) exhibit strong numerical sequence modeling capabilities and show promise in time-series prediction. While LLMs display in-context learning capabilities, the mechanisms with which they accomplish time-series prediction remain unclear. Specifically, whether they truly understand the underlying structure, which at a minimum requires reasoning over first differences in the sequence of numbers. To study this, we investigate Llama 3.1-8B from a mechanistic interpretability point of view. Mechanistic interpretability is an emerging field concerned with the reverse engineering of the algorithms learned by neural networks such as LLMs. To assess Llamas' numerical sequence modeling capabilities and to facilitate our mechanistic interpretability analysis, we create a sequence modeling task that cannot be solved without picking up structural cues. Specifically, we sample n random numbers and repeat them with an offset. We find that Llama displays strong performance on our tasks suggesting that it can pick up on the underlying structure. To understand the mechanisms that allow it to do so, we perform probing experiments and activation patching based counterfactual analysis. Probing reveals that the model computes and stores first differences in its internal representations without explicit supervision, indicating that it tracks structural information about the sequence. Activation patching reveals that Llama retrieves the relevant first-difference with a mechanism similar to an induction circuit and subsequently adds it to the current value. Notably, our work represents one of the first studies to identify this form of concept induction in LLMs.

Combinatorial scheduling poses a significant challenge for language models, requiring them to identify feasible solutions within exponentially large search spaces while satisfying complex constraints. This challenge is especially pronounced in resource-constrained settings, where larger language models are impractical and selection is limited to smaller models which often fail to preserve feasibility when scheduling directly from natural language. To address these limitations, we introduce SDDL, a neuro-symbolic framework that translates natural-language scheduling problems into compact, solver-aligned representations of tasks, resources, constraints, and objectives, while delegating low-level modeling and search to a deterministic compiler and external solver. On a 300-instance, multi-family subset of scheduling problems, SDDL improves independently verified feasibility for every resource-constrained model tested. The two strongest SDDL configurations reach 55.3% and 28.3%, up from direct-generation baselines of 23.7% and 1.3% and solver-code baselines of 21.7% and 7.0%, with a 0.0% median optimality gap among feasible schedules. By expressing problem structure rather than generating solutions or solver code, SDDL enables smaller models to approach the strongest evaluated direct- and solver-code configurations, including substantially larger frontier models.

We answer this question with Vector-Symbolic Policy Gradient (VSPG), a discrete-action actor that represents each action by a unit-norm hypervector and scores it by similarity to the encoded state. Under the standard softmax policy-gradient surrogate, we prove that its update is exactly advantage-weighted hypervector bundling followed by normalization, and therefore supports standard advantage estimators. We further show that each trained action hypervector is a fixed-size compressed kernel memory, storing an advantage-weighted kernel expansion over visited states and transferring evidence according to the encoder-induced similarity. This provides a concrete mechanism that can support sample-efficient learning without increasing inference-time memory. Finally, for bipolar action memories, we prove that greedy action selection is stable under random bit flips, with failure probability decaying exponentially in the hypervector dimension. VSPG thus connects VSA action memories, log-linear policy gradients, and kernel policy search while providing a quantitative robustness guarantee.

Large language model (LLM) agents can now carry out long-horizon technical workflows involving complex tool use, code execution, file edits, and generated artifacts. As agents do more work faster, the productivity bottleneck shifts from producing outputs to auditing whether those outputs are correct and trustworthy. Agent observability systems make fine-grained execution events visible, but visibility alone still leaves reviewers to reconstruct which actions, artifacts, and validation steps matter for a particular conclusion. We introduce LEDGER - Layered Evidence and Decision Graphs for Execution Review, a tracing and review system that builds layered trace graphs over observed agent sessions. LEDGER preserves Trace Records while grouping them into Evidence Nodes and Workflow Nodes, representing artifacts as evidence anchors, and adding typed semantic edges that connect claims to supporting actions, artifacts, and checks. Through data-analysis and coding examples, we show how the resulting traces expose workflow decisions, artifact lineage, repair steps, validation coverage, and claim-support paths for evidence-centered audit.

Wearable stress classifiers can achieve strong average performance while failing completely for a particular individual. On WESAD, a Random Forest reaches 93.0% mean accuracy yet yields F1 = 0 for Subject 14, whose cross-signal coupling weakens near stress onset. We call this structural ambiguity: individually plausible physiological channels form an inter-signal pattern that is poorly supported by the person's non-stress reference. We introduce the Individual Conformal Coupling Monitor (ICCM), a lightweight and transparent pre-inference monitor that quantifies subject-specific coupling divergence and routes each window to classify, defer, or abstain without retraining the downstream classifier. Across WESAD (N = 15) and Stress-Predict (N = 35), full-cohort Pearson associations between ambiguity and accuracy are negative (r = -0.607, p = 0.016; r = -0.412, p = 0.014). Robustness analyses temper this finding: rank correlations are not significant, and the WESAD association disappears when Subject 14 is removed. ICCM changes false-positive counts from 29 to 27 and 94 to 92, although neither paired change is significant. It withholds 3 of Subject 14's 21 stress windows but does not repair the missed-stress failure. These results position ICCM as an interpretable signal of unsupported physiology and individual failure, rather than a stand-alone safety guarantee.