Generative LLM-based recommenders (LLM-Rec) require continual post-deployment updates, yet deployment logs provide only policy-shaped contextual bandit feedback: outcomes are observed solely for items exposed by a prior serving policy, inducing exposure bias and yielding partial, asymmetric signals consisting of relatively reliable positive responses and ambiguous no-responses. We propose an Anchored Bandit Policy Optimization (ABPO) framework for continual LLM-Rec updates that combines group-relative policy optimization (GRPO) with explicit treatment of exposure bias and feedback ambiguity. Specifically, we insert the exposed recommendation as a logged anchor into each GRPO rollout group, so that group-relative normalization is calibrated against the action actually exposed by the prior policy rather than against newly sampled rollouts alone. Because both positive- and no-responses are observed only through prior-policy exposure, we apply self-normalized inverse propensity scoring to the fixed anchor for both feedback types to correct for policy mismatch. At the same time, we treat the two feedback types asymmetrically in reliability: positive responses provide relatively direct endorsement signals, whereas no-responses remain ambiguous because they may reflect either true disinterest or unobserved external factors. To avoid overly aggressive updates from ambiguous no-responses, we temper their penalties with self-certainty, using the model's output-token confidence as a verifier-free reliability signal. Across five domains from Amazon Reviews and MovieLens, our method yields consistent post-update gains in recommendation accuracy while mitigating prior-policy-induced exposure bias more effectively than prior baselines.

Safety validation of autonomous driving systems requires high-risk scenario coverage, clear collision semantics, executable trajectories, and attributable multi-vehicle interactions. Existing safety-critical scenario generation methods often rely on low-level trajectory perturbations, collision-proxy optimization, or single-adversary search, which may produce adversarial samples with ambiguous collision causes or uncontrolled multi-vehicle collisions. This paper proposes KG-ASG, a collision-knowledge-guided closed-loop adversarial scenario generation framework with primary-support attribution. KG-ASG constructs a structured collision knowledge base and trains a lightweight Collision Expert to infer the target collision mode, the unique primary adversary, support vehicles, and their interaction roles. Guided by this semantic prior, multi-vehicle adversarial generation is formulated as a primary-support process, where the primary adversary induces the main conflict and support vehicles shape the surrounding risk structure without becoming additional colliders. Rule, physical, interaction-safety, and single-collider constraints are imposed as hard gates to filter non-executable samples. To handle reactive ego behaviors, planner-controller feedback is further used for failure diagnosis, candidate re-ranking, and terminal refinement. Experiments on WOMD scenarios reconstructed in MetaDrive show that KG-ASG achieves strong adversarial effectiveness while improving Valid Primary Attack, reducing multi-collision, and obtaining closed-loop recovery gains under IDM, Cruise, and Expert controllers. These results demonstrate that collision-knowledge guidance and primary-support single-collider reasoning improve adversarial effectiveness, interpretability, and executability for autonomous driving safety validation.

Federated learning (FL) enables collaborative learning of computer vision models, where privacy and regulatory constraints prevent centralizing data across devices or organizations. However, practical FL deployments often exhibit severe class imbalance and label skew, causing standard aggregation protocols to overfit dominant clients and degrade minority-class performance. We propose a data-free, class-wise contribution estimation and aggregation framework based on logit maximization (CELM) that does not require sharing raw data, client metadata, or auxiliary public datasets. The FL server probes client updates to obtain class-wise evidence scores and assembles a cross-client evidence matrix, which quantifies both per-class competence and class coverage. Using this matrix, we compute contribution weights that upweight clients providing strong, discriminative evidence for underrepresented classes. The resulting aggregation is stable due to simplex constraints and momentum smoothing, and it remains compatible with standard FL training pipelines. We evaluate the approach on representative vision benchmarks under controlled non-IID and pathological label splits, demonstrating that CELM-based aggregation improves robustness to imbalance and statistical heterogeneity, while yielding better performance without requiring any additional data exchange.

Evaluations of unlearning on reasoning models sometimes show a bypass pattern. The answer side looks unlearned, but the model's own thinking trace keeps emitting the forgotten content, and the gap is taken as evidence that the weights still remember. We audit this reading on DeepSeek-R1-Distill-Qwen-7B with LoRA-memorized fictional authors and NPO unlearning, conditioned on a six-token canary head. On one seed, swapping the thinking trace for a short non-canary prefill on the same weights drops the answer rate by as much as the bypass gap itself, whether the prefill mimics the training template or not. On a second seed the bypass gap shrinks rather than vanishing, and the prefill swap reverses direction and brings the answer rate to ceiling. A positive parser-split bypass gap thus does not by itself identify hidden weight-level memorization, and does not rule it out either. On a different distillate the same metric flips sign because the parser cannot find the closing tag. We recommend a decode-time template swap as a cheap sanity check alongside the canonical audit.

Modern machine learning forces practitioners to choose between powerful but expensive deep networks and fast but limited classical algorithms. Here we introduce Soft Learning, a framework that maintains a library of heterogeneous specialists -- spanning linear models, tree ensembles, kernel machines, and neural networks -- and discovers provably optimal combination weights through cross-validated non-negative least squares. Soft Learning is guaranteed to match or exceed the best weighted combination of its specialists, trains over two orders of magnitude faster than deep networks on CPU alone (72-435x faster across tested configurations), provides inherent interpretability through learned weights that reveal which algorithmic paradigm best fits the data, and is future-proof: adding specialists is mathematically guaranteed to maintain or improve performance. Across 37 datasets (25 classification, 12 regression) against nine methods including CatBoost and tuned deep networks, Soft Learning ranks first on 70% of tasks, achieves the best mean rank (Friedman test, p = 1.12 x 10^-12), and is the only method to simultaneously excel at both classification and regression -- all without GPU hardware or hyperparameter tuning. These results suggest a paradigm shift from "which algorithm is best?" to "what is the provably optimal combination?" -- a question Soft Learning answers with formal guarantees for any data modality.

Multimodal emotion recognition aims to integrate text, audio, and video sources to understand human affective states. Although multimodal large language models excel at multimodal reasoning, they typically treat emotion categories as independent labels, ignoring the rich hierarchical taxonomy of human psychology. Moreover, lacking external contextual knowledge makes them highly susceptible to over-interpreting noisy cues, further complicating fine-grained emotion classification. To address these issues, we propose \textbf{HyperEmo-RAG}, a retrieval-augmented generation framework that leverages a structured emotional knowledge base. Our framework introduces two key innovations. 1) Hierarchical hyperbolic grounding. Recognizing the inherent hierarchical tree structure of emotion taxonomies, we jointly embed hierarchical emotion labels and multimodal samples into a continuous hyperbolic space (Poincaré ball) and design a hierarchical beam-search deliberation process that progressively retrieves samples from coarse to fine-grained levels. 2) Structured evidence injection. Based on the retrieved evidence, we construct an evidence graph and inject the structured knowledge as explicit cognitive context into the LLM through a Tree-Aware Attention mechanism and an EmotionGraphFormer, preserving the integrity of graph-structured information. Experiments on multiple datasets demonstrate that HyperEmo-RAG significantly outperforms existing methods.

A diffusion model trained on Hamiltonian trajectories can achieve rollout MSE near $10^{-3}$, but the standard deviation of its energy over time is between 7500 and 36000 times larger than the ground-truth energy standard deviation, indicating a failure to preserve conservation laws. This gap motivates our central question of whether neural networks can learn or select globally conserved quantities from physical trajectories. We investigate this across three Hamiltonian systems: projectile motion, pendulum, and spring-mass. We use a structured $T(v)+V(q)$ energy model, a black-box Conservation Discovery Network (CDN), a polynomial CDN, and a conditional diffusion baseline. The structured network reaches $R^2 \geq 0.9999$ against analytical energy on clean data, while the black-box CDN reaches $R^2 \geq 0.996$ when trained with temporal consistency plus a small alignment loss to analytical energy at $t=0$ ($λ_{\mathrm{align}}=0.2$). With $λ_{\mathrm{align}}=0$, CDN Pearson $R^2$ collapses on pendulum and spring-mass ($< 10^{-3}$), showing that temporal consistency alone is not enough to reliably identify the true energy. Under $1\%$ additive Gaussian noise, the CDN outperforms the structured model on the projectile and spring-mass systems, suggesting that the CDN may be more robust to noisy inputs in this setting. However, the polynomial CDN is sensitive to training configuration: it achieves $R^2=0.78$ under a short training schedule on the pendulum system, but reaches $R^2=0.9998$ with more training time and data, regardless of whether noise is added.

LLM agents exhibit a consistent tendency to over-call, invoking tools even in situations where none is needed. On the When2Call benchmark, six models from three families show high call accuracy but much lower no-call accuracy, leaving overall accuracy in the 55%-70% range. We trace this to an Intrinsic Bias Hypothesis (IBH): the call/no-call decision mapping carries an activation-independent call offset, so the model favors call even at activation parity. Using Sparse Autoencoders (SAEs), we recover behavior-aligned feature bases for the call/no_call decision, reduce them to a signed activation margin, and estimate the offset directly. Across all six models, the model is decision-neutral only when no_call activation outweighs call activation, consistent with IBH. We then causally test IBH with Adaptive Margin-Calibrated Steering (AMCS), a closed-form counter-bias shift along SAE decoder directions. Cancelling the diagnosed offset mitigates over-calling and improves overall accuracy with a negligible drop in call accuracy. Our work recasts over-calling from an empirical phenomenon into a mechanistic object amenable to causal correction. Code is available at https://github.com/SKURA502/agent-sae/.

In dynamic flow fields, various animals exhibit remarkable odor search capabilities despite relying on stochastic detections. Interestingly, there exists an optimal time window for integrating these detections that maximizes search efficiency. To understand the underlying mechanism, we investigate the navigation performance of Reinforcement Learning (RL) agents in unsteady flows under varying memory lengths and flow conditions. Without any predefined models, the agents develop a flow-assisted casting strategy and adaptively adjust both the geometry of their search trajectories and the concentration threshold for initiating casting to maximize the success rate. The agent's average speed toward the odor source exhibits a non-monotonic dependence on memory length, which can be explained by the "sector-search" model.

Rare diseases such as inborn errors of immunity (IEI) require early diagnosis to prevent end organ damage and improve quality of life. Hurdles in accessing and curating large scale electronic health record (EHR) data limit routine data driven analyses to remain on the forefront of IEI and other rare disease trends. Development of machine learning (ML) algorithms in IEI for pattern recognition as well as published methodology examining how to systematically process and integrate complex medical data is limited. Our proposed pipeline, including data curation and ML clustering algorithms, is designed to recognize novel rare disease patterns and extract IEI- associated features from a national data registry. Our methodology for EHR data formatting and processing presents the pipeline that transforms raw immunologic lab data into vectors. This is further combined with hyperparameter tuning for diseases pattern recognition via clustering. This study refines IEI feature awareness, develops data tool kits for rare disease populations analysis, and expands on transforming complex medical records in data structures interpretable by unsupervised ML.

Robotic assembly in architectural construction faces a persistent bottleneck: existing planners are either highly specialized, requiring prohibitive retraining for every new geometric design, or operationally inefficient, treating structural sequencing and kinematic motion as disjoint processes. We present EUPHORIA, a unified framework that achieves universal few-shot adaptability and dynamic efficiency through a hybrid optimization strategy. To overcome the retraining bottleneck, we propose a Meta-Geometric Encoder based on Graph Hypernetworks: unlike standard contrastive learning, which performs only feature-level recognition, our hypernetwork dynamically generates policy parameters from a minimal support set, enabling parameter-level adaptation to complex topologies (e.g., domes, arches) without gradient-based retraining. For structural reasoning, we introduce a Physics-Informed Graph Transformer trained via Soft Actor-Critic (SAC), with a Physics-Bias Attention mechanism that modulates attention scores using contact forces from Discrete Element Model (DEM) simulations, guiding the planner toward structurally critical connections. We further ensure operational efficiency through Kinematics-Aware Sequencing, where the SAC objective penalizes high-energy transitions. Finally, we bridge the Sim2Real gap via Residual Stability Correction, a differentiable optimization layer that fine-tunes coarse assembly actions by minimizing a joint energy-stability cost prior to execution. Experiments show that EUPHORIA significantly reduces energy consumption over decoupled baselines and achieves state-of-the-art success rates on unseen, non-standard geometries with minimal few-shot examples, fusing meta-learning, physics-informed attention, and residual optimization into a cohesive, generalized planner.

When Large Language Models produce structured outputs such as travel plans, code solutions, or multi-step proofs, individual reasoning steps may appear correct while the output as a whole violates budgets, fails test cases, or contradicts earlier deductions. We propose a decomposed energy function that combines a learned quality scorer with deterministic analytical constraint penalties for verifying structured LLM outputs. The quality scorer is a heterogeneous ensemble of low-rank adapters on a single frozen encoder (3% trainable parameters); the ensemble mean ranks candidates while the standard deviation quantifies epistemic uncertainty, driving a two-pass inference loop that triggers targeted regeneration or abstention. Across five benchmarks (GSM8K, MuSR, TravelPlanner, TACO, Knights & Knaves), our 149M-parameter verifier orchestrating a pool of 7-26B open generators outperforms single-shot Qwen-72B on every benchmark, matches Claude Sonnet 4.6 on MuSR (67.7% vs. 68.0%), and reduces constraint violations by 53% relative to Opus 4.6 on TravelPlanner (oracle 0.028, random 0.231). The two routes are complementary: structural verification wins when constraints are checkable (the verifier captures signal frontier models cannot self-detect), while pretraining-scale priors win where they are not (narrative inference, code semantics). A cross-dataset confounding analysis confirms genuine quality discrimination on four reasoning tasks and identifies a model-identity shortcut on code, mitigated via last-layer retraining. Scorers trained on difficult data transfer zero-shot: a MuSR-trained scorer achieves 93.9% on GSM8K without seeing a math problem.

Large language models (LLMs) achieve strong performance across a wide range of tasks but are highly sensitive to prompt design, motivating the need for automatic prompt optimization. Existing methods predominantly focus on performance alone, ignoring competing objectives such as inference cost or latency. At the same time, existing work on multi-objective prompt optimization relies on off-the-shelf NSGA-II, ignoring optimization efficiency. As a remedy, we introduce MO-CAPO, a novel multi-objective prompt optimization algorithm that jointly optimizes performance and inference cost while leveraging budget allocation for cost-efficient optimization. We further propose a deployment-oriented cost objective that captures the full computational profile of LLM inference. We evaluate our approach across four tasks and three LLMs and compare it to an NSGA-II-based multi-objective method and state-of-the-art single-objective prompt optimizers. Results show that MO-CAPO consistently identifies strong, robust, and diverse Pareto front approximations while maintaining cost-efficiency. It outperforms the NSGA-II baseline on 8 out of 12 cases in terms of the noisy R2 metric and achieves competitive performances often already at a considerably lower budget. The discovered solution sets span diverse performance-cost trade-offs that are omitted by single-objective optimizers, yet the top-performance candidates remain competitive with single-objective solutions. Additionally, we conduct the first evaluation of multi-objective machine learning experiments that considers generalization and robustness through noisy R2 and approximation gap, enabling a more realistic assessment of solution quality. MO-CAPO enables practitioners to select from an efficiently discovered set of multiple prompts offering different trade-offs between performance and cost.

Test-time model evolution offers a promising way for deployed models to improve from unlabeled test-time experience, yet most existing methods depend on backpropagation (BP), which incurs substantial memory overhead and makes them difficult to deploy on edge devices, quantized models, specialized accelerators, or black-box models. In this work, we study test-time model evolution under a strict two-forward budget, a setting that pushes adaptation toward highly efficient real-world deployment. We reveal three key obstacles in zeroth-order test-time optimization: susceptibility to shortcut solutions, uncontrolled weight drift, and ineffective update direction estimation. To overcome them, we propose EVA-0, a minimal zeroth-order adaptation framework that: 1) keeps the loss scale-invariant to prevent shortcut solutions; 2) devises an anchor-guided optimization strategy to alleviate weight drift; 3) uses sample-wise symmetric two-sided perturbation for update direction estimation and inference. EVA-0 requires no BP and performs both inference and adaptation within only two forward passes per sample. Results on ImageNet-C & ViT-Base show that EVA-0 outperforms both BP-based DeYO and BP-free FOA, while achieving a 14x speed-up over FOA. Code will be released.

Self-attention serves as the core foundation of large-scale transformer pretraining, but its quadratic token interaction cost makes inference expensive. Replacing attention with simpler sequential modules is appealing, yet naive substitution is often lossy, especially at larger scales. This paper revisits attention replacement through the lens of sparsity. Based on the observation of diverse sparsity patterns across transformer layers, we posit that pretrained transformers decompose the complex token dependency across tokens into various sequence-to-sequence mappings of diverse complexities, where some layer functionalities can be approximated and replaced with much simpler sequential modules without loss. We evaluate this premise using a plug-and-play layer-wise distillation framework to approximate and replace attention functionalities in pretrained vision transformer models. Controlled group-wise replacements under a fixed training budget reveal a clear pattern: substituting layers with sparser attention incurs substantially smaller accuracy drops than replacing denser ones. We further impose explicit attention sparsity on the pretrained ViT via AViT-style token retention and perform sparsity-guided distillation for sequential replacing models, where we see increasing teacher sparsity consistently reduces the student-teacher gap. The proposed method achieves efficient attention replacement for reduced parameter size and latency through the guidance of attention sparsity.

Recent studies observe that reinforcement learning with verifiable rewards (RLVR) reliably improves pass@1 on reasoning tasks, yet often fails to yield comparable gains in pass@k, raising the question of whether RLVR genuinely enables large language models to acquire novel reasoning abilities or merely enhances the efficiency of sampling reasoning modes already present in the base model. Prior analyses largely support the latter view, attributing this limitation to structural properties of standard RLVR objectives that result in insufficient exploration pressure. In this work, we argue that a central structural constraint arises from reverse-KL regularization, which stabilizes training but inherently anchors the policy to the reference distribution, thereby suppressing the emergence of alternative reasoning modes. However, we show that neither removing the KL term nor replacing it with forward-KL provides a satisfactory solution, as both disrupt the efficiency-coverage trade-off by either inducing reward hacking or allocating probability mass to off-target regions. To resolve this tension, we propose SAGE, a principled framework that enables controllable empirical support expansion by reshaping the reverse-KL anchor distribution itself through a guide function q(x,y), achieving consistent improvements in both pass@1 and pass@k across challenging mathematical reasoning benchmarks. Our code is available at https://github.com/tally0818/SAGE.

Cardiovascular disease remains the leading cause of death worldwide, and early detection of arrhythmias through continuous ECG monitoring on wearable devices can prevent life-threatening events. Federated Learning (FL) enables privacy-preserving collaborative training by keeping raw ECG data on device, yet standard FL incurs prohibitive communication overhead and standard deep learning models cannot fit on ultra-low-power microcontrollers. We propose Family-Grouped Hierarchical Federated Learning (Family-FL), a three-tier architecture that uses the family as a natural privacy boundary for intra-family aggregation before global synchronization. We further design a hardware-constrained Tiny CNN-LSTM architecture with only 669 parameters, INT8-quantized to occupy merely 4.65KB Flash and 2.95KB RAM, meeting the constraints of STC32G12K128-class microcontrollers. Experiments on the MIT-BIH Arrhythmia Database (mean of 5 independent runs with different seeds) demonstrate that Family-FL reduces communication volume by 76.7% compared to FedAvg while maintaining comparable accuracy. Family-FL-Tiny achieves 91.9 +/- 1.2% accuracy with macro-F1 of 0.483 +/- 0.031, reducing total communication to 0.31% of FedAvg. The model achieves reliable ventricular arrhythmia detection (per-class F1 = 0.80), the most clinically critical abnormality for home-based preliminary screening. These results demonstrate the technical feasibility of privacy-preserving federated learning on ultra-resource-constrained microcontrollers through simulation-based evaluation. We honestly discuss limitations: no hardware deployment, single-dataset validation (MIT-BIH, 47 subjects), reduced rare-class sensitivity, and absence of formal differential privacy guarantees.

Probabilistic prediction systems often aggregate probability estimates from multiple models into a single decision. A common assumption is that if each model is individually calibrated, the aggregate prediction will also be well calibrated. We show that this assumption fails in multi-agent settings: individually calibrated predictors can become collectively miscalibrated when their predictions interact strategically, in the game-theoretic sense of Brier-optimal local response, even without deliberate coordination. This phenomenon arises naturally when agents are independently trained on overlapping data. We prove that under Brier-score-based aggregation with positively correlated beliefs, each agent's individually optimal report systematically underestimates the positive-class probability, yielding a Price of Anarchy greater than one whenever Cov(b_i, b_j) > 0. In a canonical setting (n = 5 agents, pairwise correlation = 0.5, base rate = 0.3), the empirically measured PoA in false-negative rate reaches 7.25x. In contrast, VCG-based aggregation aligns incentives by rewarding marginal contribution, achieving dominant-strategy incentive compatibility and near-optimal performance. Experiments on three real-world datasets (NSL-KDD, UNSW-NB15, Credit Card Fraud) show that VCG provides strong robustness while maintaining comparable accuracy. It performs particularly well in data-sparse and adversarial settings, and adaptive weighting further improves performance under distribution shift.

Attention Residuals replace standard additive residual connections with learned softmax attention over previous layer outputs, enabling selective cross-layer routing. However, standard Attention Residuals still attend over cumulative hidden states in previous layers, which are highly redundant. We show that this redundancy leads to routing collapse in deeper layers: attention weights become low-contrast and closer to uniform (max weight ${\approx}$0.2), limiting the model's ability to select informative states in previous layers. This raises a key but underexplored design question: what layer-wise representations should be routed in Attention Residuals? To answer this question, we propose Delta Attention Residuals, which attend over deltas -- the change introduced by each sublayer ($\mathbf{v}_i = \mathbf{h}_{i+1} - \mathbf{h}_i$) -- instead of cumulative states. Delta representations are structurally diverse and yield higher-contrast attention distributions (max weight ${\approx}$0.6), enabling more selective and effective routing across layers. This principle applies at both per-sublayer and block granularity. Across all tested scales (220M--7.6B), Delta Attention Residuals consistently outperform both standard residuals and Attention Residuals, with 1.7--8.2\% validation perplexity gains. Delta Attention Residuals also enables converting pretrained checkpoints into Delta Attention Residuals via standard fine-tuning. Code is available at https://github.com/wdlctc/delta-attention-residuals-code.

Coding agents accumulate extensive context during long-running tasks, yet fixed context windows force practitioners to choose between truncation and task failure. While numerous memory condensation strategies have been proposed, from simple sliding windows to LLM-generated summaries, no systematic comparison exists to guide strategy selection, especially in scientific discovery tasks. We evaluate eight memory condensation strategies using GPT-4o on sixty DiscoveryBench tasks spanning six scientific domains (480 total evaluations). We find that no condenser significantly alters hypothesis quality, while LLM-based condensers increase token costs by 24-94 percent, and masking tool-call outputs achieves an 8.6 percent net savings. We also observe that the optimal condenser for data-driven scientific discovery varies by scientific domain and task length.

Edge deployment of Vision-Language Models (VLMs) faces a tradeoff between latency and accuracy: cloud execution provides high-quality predictions but incurs communication delay and energy cost, while edge-only execution is faster but less accurate due to limited model capacity. This trade-off is further complicated by heterogeneity in image quality and reasoning complexity, making static placement suboptimal. We present INAR-VL, a lightweight edge-cloud routing system for multimodal inference in a two-tier deployment. INAR-VL maintains complementary VLMs across edge and cloud and uses lightweight image and text complexity signals to guide routing and model selection, executing simple queries locally while offloading complex ones when beneficial. Evaluation on visual question answering shows that INAR-VL executes 36% of requests on the edge, reduces latency by 24%, lowers energy by 26%, and preserves 97% of cloud-level accuracy.

Checkpoint selection for multimodal large language models (MLLMs) presents significant challenges when performance differentials are marginal and evaluation signals are prone to noise. Existing methodologies rely heavily on static benchmarks or pointwise scoring, which frequently misalign with in-the-wild usage and lack robust uncertainty estimation, particularly in OCR-heavy scenarios. In this work, we formulate checkpoint selection as a robust decision problem under evaluation uncertainty. We propose a multi-stage framework that integrates curated real-world data, structured LLM-based judgment, and multi-stage ranking protocols. The evaluation system orchestrates progressive refinement via pointwise filtering, listwise ranking, and pairwise comparison. To enhance reliability, we introduce subsampling-based confidence estimation and a percentile-based scoring formulation that captures distributional characteristics while penalizing tail failures. Furthermore, we demonstrate that data quality, specifically OCR readability, is a critical determinant of evaluation validity.

Recent advances in Reinforcement Learning (RL) have underscored its potential for incentivizing reasoning capabilities of Large Language Models (LLMs). However, existing step-level efforts suffer from costly annotations that limit domain coverage, while scalar scores further impose an information bottleneck, offering insufficient semantic bandwidth to improve intermediate decisions. Alternative language-critique approaches, which rely on frozen or external critics, provide richer textual feedback but lack the scalability needed for sustained policy improvement. In this work, we propose language-driven stepwise trajectory redirection, termed as STRIDE, a novel training framework that shifts process supervision from scalar rewards to learnable stepwise language feedback. Specifically, we co-train a generator and a generative verifier using only outcome-based rewards, eliminating external annotations, while delivering sustained policy improvement through jointly aligned verifier training. The verifier's stepwise language critiques explicitly localize and explain failures, enabling the generator to redirect reasoning trajectories at intermediate steps toward alternative decisions. The trajectory redirection design guarantees harmless policy improvement, even under noisy or suboptimal verifier feedback. Experiments on diverse reasoning benchmarks show that STRIDE significantly outperforms state-of-the-art baselines, as well as achieving breakthroughs on zero-pass-rate problems where scalar methods yield no learning signal in our ablation studies, demonstrating the effectiveness of learnable stepwise language feedback for enhancing LLM reasoning.

Explainability methods have progressed rapidly, but global explanations for time-series models remain underdeveloped, with most approaches focusing on local, instance-level attributions. We introduce INSIGHTS, a model-agnostic, user-centric approach for providing global explanations of time series models. Our approach prioritizes simplicity, efficiency, and transparency in its design, ensuring that stakeholders can readily adopt its outputs. While current methods focus on local explanations, INSIGHTS generates sample summaries that offer a comprehensive overview of model behavior. It balances the importance and diversity of time series samples to create informative subsets using utility functions that capture domain-specific aspects of time series behavior, such as exceeding domain norms. We evaluate INSIGHTS through experiments, interviews, and a user study. Our results indicate INSIGHTS effectively constructs comprehensive, diverse time series subsets, producing summaries manageable for individual evaluation. It is preferred by domain experts for its ability to provide a stable understanding of model behavior and the quality of the samples identified. Moreover, user study participants presented with INSIGHTS-based summaries exhibit an enhanced understanding of the model's overall behavior.

Do transformers, when trained on sequential reasoning traces, build internal models of the underlying task? And if so, does the structure of those internal representations mirror the structure of the domain? We train an 8-layer transformer on Sudoku solving traces and perform a mechanistic analysis of its internal computation. We establish two results. First, the model builds a substructure world model: it does not represent the board state cell by cell, as a human analyst would expect, but organizes information around the rows, columns, and boxes that Sudoku's constraints act on. Second, we identify a naked-single circuit: a small set of dedicated neurons in the final MLP layer, each individually detecting when exactly one digit remains possible for a specific cell, and reliably promoting that digit. These findings show that the geometry of an emergent world model is shaped by the constraint algebra of the domain, not its surface presentation, and that the resulting decision circuit is sparse, monosemantic, and fully interpretable. More broadly, they demonstrate that mechanistic interpretability tools can recover an end-to-end algorithmic account of how a transformer solves a combinatorial reasoning task.

Classical communication systems fail not only through random noise but also when transmitter and receiver use incompatible operational codebooks. Variational autoencoders (VAEs) train an encoder $q_ϕ$ and decoder $p_θ$ jointly, and practitioners treat the resulting latent space as a discrete code -- for clustering, conditional generation, and mechanistic interpretability. Yet standard VAE diagnostics -- ELBO, active units, mutual information, and code histograms -- certify only whether this code is used, never whether the decoder reads each latent under the encoder's code. We close this gap with the neural codebook channel $K_{e\to d}(j\mid i)$, a coupled encoder-decoder diagnostic whose off-diagonal mass is bounded by an architecture-free Bernoulli-KL certificate $d_{\mathrm{bin}}(1-\mathcal{A} \,\|\, \barη_p) \le \barΔ$ controlled by the variational gap. The certificate is the operational specialization of the classical KL chain rule under disintegration to the encoder-decoder disagreement event, complemented by a constructive marginal-impossibility result: no combination of marginal histograms, entropies, active-code counts, or mutual information determines $K_{e\to d}$. We audit the certificate on four sklearn datasets (finite-grid exact, 5/5 seeds, 20/20 pairs satisfy the bound), a 2D model where the bound is non-vacuous at $2.71\times$ the observed disagreement and the four-term identity closes within $10^{-4}$, MNIST under importance-sampling control, and a VQ-VAE attaining the predicted limit $\hat{\mathcal{A}}=1.000$. The package $(K_{e\to d}, \mathcal{A}, R_{\mathrm{eff}}, R, \mathrm{AU})$ is an audit-ready reporting unit. More broadly, the framework makes mismatched decoding -- a failure mode classical communication theory named decades ago -- visible inside a single deep generative model.

We give the first quantitative prediction of grokking delay under AdamW. Treating the delay as a first-passage time, we derive a closed-form law T_grok - T_mem = (1 / 2 kappa_LL eta lambda) log(V_mem / V_star), where V_t = ||theta_t||^2 is the squared parameter norm, V_star is an architecture-dependent threshold, and kappa_LL absorbs the AdamW correction to the clean-SGD contraction rate 2 eta lambda. Calibrating (kappa_LL, V_star) on a single hyperparameter cell predicts grokking delays on 26 held-out runs with MAPE 17.7% over a 41x delay range; the law generalises to MLPs (MAPE 18.0%, N=34) and degrades to 23.3% on cross-task extension (N=46, 43.5x range), with a structured residual in which V_star / V_mem stays comparatively stable within architecture (CV about 14% on the 1L transformer). First-passage of V_t is necessary but not sufficient. A quantile-margin theorem establishes that positive delay requires both norm separation V_mem > V_post and angular reachability of a threshold alpha_star = arcsin(C / V_T_mem^(1/2)), where C is computable from the empirical NTK feature map and the validation-margin quantile. Calibrating C on modulus p=89 predicts alpha_star = 47.2 degrees at p=97 (observed 47.8 degrees, error 1.3%) as a prior cross-cell prediction. Causal interventions that freeze the norm or remove weight decay at memorisation eliminate grokking (0/6 vs. 3/3 baseline), trapping the angular displacement near 12 degrees. kappa_LL is empirically measured per architecture rather than derived from (beta_1, beta_2, epsilon); within-architecture CV stays at most 15% across four architectures, but values differ by about 2x between architectural variants beyond depth alone. Empirical scope is algorithmic tasks (modular arithmetic, sparse parity) under AdamW; whether the law transfers to natural-language scale models is open.

With the deep integration of the travel and energy industries, cross-industry supply chain finance has gradually become a high-risk field of hidden money laundering incidents. For this reason, this work proposes a graph-driven cross-industry real-time anti-money laundering monitoring framework (GCRMF) for integrated travel - energy supply chain networks. First, a cross-industry heterogeneous graph (CIHG) covering new energy vehicle rental platforms, energy suppliers, fintech institutions, etc., is constructed, and industry semantics are integrated through temporarily Dual-GAT (Temporal Dual-Graph Attention Network), dynamically encoding capital flow paths and evolution features over time. Subsequently, in order to identify the structural fraud behavior together produced by colluding subjects, a meta-path subgraph reasoning module based on contrastive learning and hierarchical graph sampling is proposed to enhance the discrimination capability of cross-industry recurring money laundering behavior. Meanwhile, a self-supervised online learning mechanism is adopted for real-time adaptation and continuous optimization to new money laundering strategies. The experimental results show that compared with existing graph neural network methods in cross-industry scenarios, GCRMF improves the performance by more than 17.8% of F1 score and greatly reduces the false positive rate.

Backtesting large language models on historical events requires reasoning exclusively from information available before a specified cutoff date. Yet models routinely leak post-cutoff knowledge from pre-training into their reasoning, inflating apparent accuracy and undermining evaluation validity. Prompt-based constraints fail when suppressed content is causally related to the prediction, and knowledge unlearning cannot address this problem because temporal compliance is instance-specific: the same fact may be legitimate evidence for one cutoff date and a violation for another. Rather than erasing knowledge, the model must learn temporal discipline: selecting evidence conditioned on each instance's cutoff date. We propose TEMPO (Temporal Enforcement via Mode-separated Policy Optimization), which trains this discipline via two contributions: (1) a two-mode reward where a leakage mode drives post-cutoff claims to zero as a hard prerequisite before a performance mode optimizes task performance; and (2) a GRPO-based training pipeline that enables the model to discover temporally valid reasoning strategies. We prove that training monotonically decreases leakage, converges to the leak-free optimum, and improves task performance once compliance is achieved. On three prediction tasks and two models, TEMPO reduces leakage from 2~13% to 0.6~3.7% across all conditions, with task performance improving 6~13% where strong pre-cutoff signals exist and maintained where the prediction task is inherently difficult from valid information alone.

Safe reinforcement learning in nonstationary environments requires safety mechanisms that adapt as environmental conditions change. Standard safe reinforcement learning methods often assume fixed constraints or stable environmental conditions, which can become inadequate under distribution shift. We propose LILAC+, a framework for safe continual reinforcement learning under nonstationarity that combines three adaptive safety mechanisms: context-based safety constraints, adaptation-speed constraints, and budget-to-state safety enforcement. Context-based constraints adjust safety requirements using inferred and predicted environmental context. Adaptation-speed constraints tighten safety requirements when the rate of environmental change exceeds the agent's ability to adapt safely. Budget-to-state enforcement converts cumulative safety requirements into local state-level control constraints that can be enforced at decision time. Together, these mechanisms provide a unified approach for proactive and reactive safety adaptation in continual reinforcement learning. We evaluate the framework in simulated driving environments under stationary, seen nonstationary, and unseen nonstationary conditions. The results show that adaptive safety constraints substantially reduce safety violations under distribution shift while maintaining competitive task performance compared with unconstrained and fixed-constraint baselines. These findings suggest that safe continual reinforcement learning requires adaptive constraint mechanisms that respond not only to current state information but also to predicted environmental context, adaptation demand, and remaining safety budget.