Search arXiv⌕ Search

arXiv subjects

Ming Jin

Publications and source records attributed to Ming Jin.

At least 55 records · Page 3Linked to original sources

Remembering More, Risking More: Longitudinal Safety Risks in Memory-Equipped LLM Agents

Safety evaluations of memory-equipped LLM agents typically measure within-task safety: whether an agent completes a single scenario safely, often under adversarial conditions such as prompt injection or memory poisoning. In deployment, however, a single agent serves many independent tasks over a long horizon, and memory accumulated during earlier tasks can affect behavior on later, unrelated ones. Studying this regime requires evaluation along the temporal dimension across tasks: not whether an agent is safe at any single memory state, but how its safety profile changes as memory accumulates across many independent interactions. We call this failure mode temporal memory contamination. To isolate memory exposure from stream non-stationarity, we introduce a trigger-probe protocol that evaluates a fixed probe set against read-only memory snapshots at varying prefix lengths, together with a NullMemory counterfactual baseline for identifying memory-induced violations. We apply this protocol across three deployment scenarios spanning records, memos, forms, and email correspondence and eight memory architectures, and additionally on Claw-like AI agents, such as OpenClaw, using the platform's native memory mechanism. Memory-enabled agents consistently exceed the NullMemory baseline, and memory-induced violation rates show a robust upward trend with exposure length on both agent classes. Order-randomization experiments indicate that the effect is driven primarily by accumulated content rather than encounter order. Finally, a structural consequence of the event decomposition is that memory-induced risk is detectable from retrieval state before generation, which we confirm with a high-recall diagnostic monitor. Our results argue for treating memory safety as a longitudinal property that requires temporal evaluation, not a single-state property that can be captured by a snapshot.

cs.AI↗

How Do Electrocardiogram Models Scale?

While scaling laws have established a fundamental framework for foundation models in natural language processing, their applicability to electrocardiogram (ECG) models remains poorly characterized. Indeed, recent studies do not always yield consistent downstream gains as one increases the model size or pre-training dataset size of ECG models, leaving the exact roles of architectural inductive biases, pre-training paradigms, and expected improvements with size largely unanswered. In this work, we systematically investigate neural and loss-to-loss scaling laws within the ECG domain. By pre-training over $120$ models (ranging from $20$K to $200$M parameters) on the large-scale CODE dataset ($2.3$M records), we decouple the effects of model architecture (ResNet vs. Transformer) and pre-training paradigm, namely supervised learning (SL) versus self-supervised learning (SSL). We found that (i) SL models are data-bottlenecked in-distribution, whereas SSL models scale robustly across both model and data sizes; (ii) for out-of-distribution (OOD) generalization, ResNets are $1.3$ to $2.5$ times more parameter-efficient than Transformers, while SSL is up to $16$ times more data-efficient and achieves up to $7.6$ times higher transfer efficiency than SL on unseen clinical tasks; (iii) across the observed scales, ResNet-based models generally achieve the lowest OOD loss, with SSL dominating on unseen clinical tasks and self-supervised Transformers overtaking at very large model sizes. Our results suggest that the path to effective ECG foundation models lies in the strategic alignment of architecture and paradigm rather than brute-force scaling.

cs.LG↗

LLMs Should Express Uncertainty Explicitly

Large language models (LLMs) often produce confident yet incorrect answers, which can lead to risky failures in real-world applications. We study whether post-training can make a model's self-assessment explicit: when the model is uncertain, can it be trained to signal so within its own response? A central design question is where in the response this signal should be exposed -- during reasoning, while the answer is still being formed, or at the end, once the answer has been produced. We study both. For end-of-reasoning self-assessment, we train the model to verbalize a confidence score for its response, with the aim of high confidence on correct answers and low confidence on incorrect ones. For during-reasoning self-assessment, we train the model to emit the marker whenever its current reasoning state appears unreliable. Across factual reasoning tasks, both forms sharply reduce overconfident errors while improving answer quality, and both can be used as triggers for retrieval augmented generation (RAG) to improve the final response. We further analyze their internal mechanisms: end-of-reasoning verbalized confidence sharpens a confidence-related structure already present in the pretrained model, whereas during-reasoning emission teaches the model to mark high-risk reasoning steps, with parameter changes concentrated in the model's late layers.

cs.LG↗

LeapTS: Rethinking Time Series Forecasting as Adaptive Multi-Horizon Scheduling

Time series forecasting serves as an essential tool for many real-world applications, supporting tasks such as resource optimization and decision-making. Despite significant architectural advancements, most modern models still treat forecasting task as a fixed mapping from history to target horizons. This induces temporal decoupling across future time points and limits the model's ability to adapt to the evolving context as forecasting progresses. In this work, we present LeapTS, a novel framework that reformulates time series forecasting as a dynamic scheduling process over the prediction horizon. Specifically, LeapTS organizes the forecasting process into multi-level decisions using: (1) the hierarchical controller to dynamically select the optimal prediction scale and advancement length at each step, and (2) continuous-time state evolution driven by neural controlled differential equations. Within this process, the controlled update mechanism explicitly couples the irregular temporal dynamics with discrete scheduling feedback. Extensive evaluations on both real-world and synthetic datasets demonstrate that LeapTS improves overall forecasting performance by at least 7.4% while achieving a 2.6$\times$ to 5.3$\times$ inference speedup over representative Transformer-based models. Furthermore, by explicitly tracing the scheduling trajectories, we reveal how the model autonomously adapts its forecasting behavior to capture non-stationary dynamics.

cs.LG↗

EventTSF: Event-Aware Non-Stationary Time Series Forecasting

Time series forecasting is vital in diverse sectors such as energy and transportation, where non-stationary dynamics are deeply intertwined with external events in other modalities such as texts. However, incorporating natural language-based external events to improve non-stationary forecasting remains largely unexplored, as most approaches still rely on a single modality, resulting in limited contextual knowledge and model underperformance. Enabling fine-grained multimodal interactions between temporal and textual data is challenged by two fundamental issues: (1) the gap in modeling interactions among discrete external events and continuous time series in a unified framework; (2) classical uniform diffusion timestep ignores event-induced non-stationary variability, leading to imbalanced denoising difficulty across diffusion stages. In this work, we propose event-aware non-stationary time series forecasting (EventTSF), an autoregressive diffusion framework that integrates historical time series and textual events via step-wise diffusion. To mitigate the imbalanced denoising difficulty of uniform timestep sampling, EventTSF uses an event-aware flow-matching timestep conditioned on event semantics. Extensive experiments on 7 synthetic and real-world datasets show that EventTSF outperforms 12 non-stationary time series forecasting baselines, achieving average gains of 41.3% in probabilistic forecasting and 27.5% in deterministic forecasting across all evaluation metrics.

cs.LG↗

The Procrustean Bed of Time Series: The Optimization Bias in Point-wise Loss Functions

Intuitively, a more deterministic time series should be easier to forecast. However, point-wise loss functions (e.g., MSE and MAE), serving as differentiable surrogates for the ideal optimization target, score each timestamp independently and therefore disregard temporal dependence. This mismatch induces a systematic optimization bias that cannot be eliminated merely by improving model expressiveness or optimizer. To formalize this issue, we define the Expectation of Optimization Bias (EOB) as the Kullback--Leibler divergence between the true joint distribution and the factorized i.i.d. surrogate induced by the point-wise paradigm. Under covariance-stationary Gaussian assumptions, we derive closed-form expressions for the stochastic component of EOB, establishing it as an irreducible lower bound on the total bias in linear systems, and further extend it to nonlinear regimes through a Gaussian mixture model lower bound. Crucially, we prove this bias is governed intrinsically by two data properties, i.e., sequence length and Structural Signal-to-Noise Ratio (SSNR), regardless of specific model architecture, optimizer, or point-wise loss forms. This theory motivates a principled debiasing program based on sequence length reduction and structural orthogonalization, which we instantiate through DFT/DWT combined with a novel harmonized $\ell_p$ norm. Extensive experiments validate the predicted SSNR--horizon dynamics, resolve the classic trigonometric fitting failure as an objective-induced pathology, and demonstrate substantial plug-and-play gains. Notably, on iTransformer, our proposed objective reduces average MSE/MAE by 5.2%/5.0% in forecasting across 11 datasets and by 27.4%/19.4% in imputation across 9 datasets.

cs.LG↗

Reinforcement Learning with Backtracking Feedback

Addressing the critical need for robust safety in Large Language Models (LLMs), particularly against adversarial attacks and in-distribution errors, we introduce Reinforcement Learning with Backtracking Feedback (RLBF). This framework advances upon prior methods, such as BSAFE, by primarily leveraging a Reinforcement Learning (RL) stage where models learn to dynamically correct their own generation errors. Through RL with critic feedback on the model's live outputs, LLMs are trained to identify and recover from their actual, emergent safety violations by emitting an efficient "backtrack by x tokens" signal, then continuing generation autoregressively. This RL process is crucial for instilling resilience against sophisticated adversarial strategies, including middle filling, Greedy Coordinate Gradient (GCG) attacks, and decoding parameter manipulations. To further support the acquisition of this backtracking capability, we also propose an enhanced Supervised Fine-Tuning (SFT) data generation strategy (BSAFE+). This method improves upon previous data creation techniques by injecting violations into coherent, originally safe text, providing more effective initial training for the backtracking mechanism. Comprehensive empirical evaluations demonstrate that RLBF significantly reduces attack success rates across diverse benchmarks and model scales, achieving superior safety outcomes while critically preserving foundational model utility.

cs.LG↗

STReasoner: Empowering LLMs for Spatio-Temporal Reasoning in Time Series via Spatial-Aware Reinforcement Learning

Spatio-temporal reasoning in time series involves the explicit synthesis of temporal dynamics, spatial dependencies, and textual context. This capability is vital for high-stakes decision-making in systems such as traffic networks, power grids, and disease propagation. However, the field remains underdeveloped because most existing works prioritize predictive accuracy over reasoning. To address the gap, we introduce ST-Bench, a benchmark consisting of four core tasks, including etiological reasoning, entity identification, correlation reasoning, and in-context forecasting, developed via a network SDE-based multi-agent data synthesis pipeline. We then propose STReasoner, which empowers LLM to integrate time series, graph structure, and text for explicit reasoning. To promote spatially grounded logic, we introduce S-GRPO, a reinforcement learning algorithm that rewards performance gains specifically attributable to spatial information. Experiments show that STReasoner achieves average accuracy gains between 17% and 135% at only 0.004X the cost of proprietary models and generalizes robustly to real-world data.

cs.CL↗

What Makes a Good AI Review? Concern-Level Diagnostics for AI Peer Review

Evaluating AI-generated reviews by verdict agreement is widely recognized as insufficient, yet current alternatives rarely audit which concerns a system identifies, how it prioritizes them, or whether those priorities align with the review rationale that shaped the final assessment. We propose concern alignment, a diagnostic framework that evaluates AI reviews at the concern level rather than only at the verdict level. The framework's core data structure is the match graph, a bipartite alignment between official and AI-generated concerns annotated with match type, severity, and post-rebuttal treatment. From this artifact we derive an evaluation ladder that moves from binary accuracy to concern detection, verdict-stratified behavior, decision-aware calibration, and rebuttal-aware decomposition. In a pilot study of four public AI review systems evaluated in six configurations, concern-level analysis suggests that detection alone does not determine review quality; calibration is often the binding constraint. Systems detect non-trivial fractions of official concerns yet most mark 25--55% of concerns on accepted papers as decisive, where, under our operationalization, no official concern on accepted papers was treated as a decisive blocker. Identical overall verdict accuracy can conceal reject-heavy behavior versus low-recall profiles, and low full-review false decisive rates can partly reflect concern dilution rather than calibrated prioritization. Most systems do not emit a native accept/reject, and inferring it from review tone is method-sensitive, reinforcing the need for concern-level diagnostics that remain stable across inference choices. The contribution is a reusable evaluation framework for auditing which concerns AI reviewers identify, how they weight them, and whether those priorities align with the review rationale that informed the paper's final assessment.

cs.AI↗

StyleBench: Evaluating thinking styles in Large Language Models

Structured reasoning can improve the inference performance of large language models (LLMs), but it also introduces computational cost and control constraints. When additional reasoning structure helps, and when it instead reduces efficiency or robustness, remains poorly understood. We propose StyleBench, where we study reasoning structure as a capacity-constrained design choice rather than a fixed inference recipe. We evaluate five representative reasoning styles: Chain-of-Thought, Tree-of-Thought, Algorithm-of-Thought, Sketch-of-Thought, and Chain-of-Draft across five reasoning tasks and 15 open-source LLMs ranging from 270M to 120B parameters. We find that greater structural complexity improves accuracy only in limited regimes defined by task demands and model capacity. Search-based styles help on open-ended combinatorial problems but fail on smaller models, while concise styles achieve large efficiency gains on structured tasks without sacrificing performance. We also identify systematic failure modes in smaller models, including premature guessing and weak adherence to reasoning-control instructions. To study adaptive reasoning control, we further compare supervised and reinforcement-based strategy selection on Qwen-7B-Instruct. Supervised fine-tuning collapses to shallow style preferences, whereas GRPO learns stronger adaptive control and improves downstream performance. Together, these results clarify when structured reasoning is useful, when it is wasteful, and why learning to choose a reasoning strategy is itself a challenging inference problem, we open source the benchmark in https://github.com/JamesJunyuGuo/Style_Bench.

cs.LG↗

Trojan-Speak: Bypassing Constitutional Classifiers with No Jailbreak Tax via Adversarial Finetuning

Fine-tuning APIs offered by major AI providers create new attack surfaces where adversaries can bypass safety measures through targeted fine-tuning. We introduce Trojan-Speak, an adversarial fine-tuning method that bypasses Anthropic's Constitutional Classifiers. Our approach uses curriculum learning combined with GRPO-based hybrid reinforcement learning to teach models a communication protocol that evades LLM-based content classification. Crucially, while prior adversarial fine-tuning approaches report more than 25% capability degradation on reasoning benchmarks, Trojan-Speak incurs less than 5% degradation while achieving 99+% classifier evasion for models with 14B+ parameters. We demonstrate that fine-tuned models can provide detailed responses to expert-level CBRN (Chemical, Biological, Radiological, and Nuclear) queries from Anthropic's Constitutional Classifiers bug-bounty program. Our findings reveal that LLM-based content classifiers alone are insufficient for preventing dangerous information disclosure when adversaries have fine-tuning access, and we show that activation-level probes can substantially improve robustness to such attacks.

cs.CR↗

FSMC-Pose: Frequency and Spatial Fusion with Multiscale Self-calibration for Cattle Mounting Pose Estimation

Mounting posture is an important visual indicator of estrus in dairy cattle. However, achieving reliable mounting pose estimation in real-world environments remains challenging due to cluttered backgrounds and frequent inter-animal occlusion. We present FSMC-Pose, a top-down framework that integrates a lightweight frequency-spatial fusion backbone, CattleMountNet, and a multiscale self-calibration head, SC2Head. Specifically, we design two algorithmic components for CattleMountNet: the Spatial Frequency Enhancement Block (SFEBlock) and the Receptive Aggregation Block (RABlock). SFEBlock separates cattle from cluttered backgrounds, while RABlock captures multiscale contextual information. The Spatial-Channel Self-Calibration Head (SC2Head) attends to spatial and channel dependencies and introduces a self-calibration branch to mitigate structural misalignment under inter-animal overlap. We construct a mounting dataset, MOUNT-Cattle, covering 1176 mounting instances, which follows the COCO format and supports drop-in training across pose estimation models. Using a comprehensive dataset that combines MOUNT-Cattle with the public NWAFU-Cattle dataset, FSMC-Pose achieves higher accuracy than strong baselines, with markedly lower computational and parameter costs, while maintaining real-time inference on commodity GPUs. Extensive experiments and qualitative analyses show that FSMC-Pose effectively captures and estimates cattle mounting pose in complex and cluttered environments. Dataset and code are available at https://github.com/elianafang/FSMC-Pose.

cs.CV↗

Amplification Effects in Test-Time Reinforcement Learning: Safety and Reasoning Vulnerabilities

Test-time training (TTT) has recently emerged as a promising method to improve the reasoning abilities of large language models (LLMs), in which the model directly learns from test data without access to labels. However, this reliance on test data also makes TTT methods vulnerable to harmful prompt injections. In this paper, we investigate safety vulnerabilities of TTT methods, where we study a representative self-consistency-based test-time learning method: test-time reinforcement learning (TTRL), a recent TTT method that improves LLM reasoning by rewarding self-consistency using majority vote as a reward signal. We show that harmful prompt injection during TTRL amplifies the model's existing behaviors, i.e., safety amplification when the base model is relatively safe, and harmfulness amplification when it is vulnerable to the injected data. In both cases, there is a decline in reasoning ability, which we refer to as the reasoning tax. We also show that TTT methods such as TTRL can be exploited adversarially using specially designed "HarmInject" prompts to force the model to answer jailbreak and reasoning queries together, resulting in stronger harmfulness amplification. Overall, our results highlight that TTT methods that enhance LLM reasoning by promoting self-consistency can lead to amplification behaviors and reasoning degradation, highlighting the need for safer TTT methods.

cs.LG↗

Parameter Stress Analysis in Reinforcement Learning: Applying Synaptic Filtering to Policy Networks

This paper explores reinforcement learning (RL) policy robustness by systematically analyzing network parameters under internal and external stresses. \textcolor{black}{We apply synaptic filtering methods using high-pass, low-pass, and pulse-wave filters from} \citep{pravin2024fragility}, as an internal stress by selectively perturbing parameters, while adversarial attacks apply external stress through modified agent observations. This dual approach enables the classification of parameters as \textit{fragile}, \textit{robust}, or \textit{antifragile}, based on their influence on policy performance in clean and adversarial settings. Parameter scores are defined to quantify these characteristics, and the framework is validated on proximal policy optimization (PPO)-trained agents in Mujoco continuous control environments. The results highlight the presence of antifragile parameters that enhance policy performance under stress, demonstrating the potential of targeted filtering techniques to improve RL policy adaptability. These insights provide a foundation for future advancements in the design of robust and antifragile RL systems.

cs.LG↗

Adversarial Déjà Vu: Jailbreak Dictionary Learning for Stronger Generalization to Unseen Attacks

Large language models remain vulnerable to jailbreak attacks that bypass safety guardrails to elicit harmful outputs. Defending against novel jailbreaks represents a critical challenge in AI safety. Adversarial training -- designed to make models robust against worst-case perturbations -- has been the dominant paradigm for adversarial robustness. However, due to optimization challenges and difficulties in defining realistic threat models, adversarial training methods often fail on newly developed jailbreaks in practice. This paper proposes a new paradigm for improving robustness against unseen jailbreaks, centered on the Adversarial Déjà Vu hypothesis: novel jailbreaks are not fundamentally new, but largely recombinations of adversarial skills from previous attacks. We study this hypothesis through a large-scale analysis of 32 attack papers published over two years. Using an automated pipeline, we extract and compress adversarial skills into a sparse dictionary of primitives, with LLMs generating human-readable descriptions. Our analysis reveals that unseen attacks can be effectively explained as sparse compositions of earlier skills, with explanatory power increasing monotonically as skill coverage grows. Guided by this insight, we introduce Adversarial Skill Compositional Training (ASCoT), which trains on diverse compositions of skill primitives rather than isolated attack instances. ASCoT substantially improves robustness to unseen attacks, including multi-turn jailbreaks, while maintaining low over-refusal rates. We also demonstrate that expanding adversarial skill coverage, not just data scale, is key to defending against novel attacks. \textcolor{red}{\textbf{Warning: This paper contains content that may be harmful or offensive in nature.

cs.LG↗

TimeOmni-1: Incentivizing Complex Reasoning with Time Series in Large Language Models

Recent advances in multimodal time series learning underscore a paradigm shift from analytics centered on basic patterns toward advanced time series understanding and reasoning. However, existing multimodal time series datasets mostly remain at the level of surface alignment and question answering, without reaching the depth of genuine reasoning. The absence of well-defined tasks that genuinely require time series reasoning, along with the scarcity of high-quality data, has limited progress in building practical time series reasoning models (TSRMs). To this end, we introduce Time Series Reasoning Suite (TSR-Suite), which formalizes four atomic tasks that span three fundamental capabilities for reasoning with time series: (1) perception, acquired through scenario understanding and causality discovery; (2) extrapolation, realized via event-aware forecasting; and (3) decision-making, developed through deliberation over perception and extrapolation. TSR-Suite is the first comprehensive time series reasoning suite that supports not only thorough evaluation but also the data pipeline and training of TSRMs. It contains more than 23K samples, of which 2.3K are carefully curated through a human-guided hierarchical annotation process. Building on this foundation, we introduce TimeOmni-1, the first unified reasoning model designed to address diverse real-world problems demanding time series reasoning. The model is trained in multiple stages, integrating a mixture of task scenarios, novel reward functions, and tailored optimizations. Experiments show that TimeOmni-1 delivers strong out-of-distribution generalization across all tasks and achieves a high rate of valid responses. It significantly improves causality discovery accuracy (64.0% vs. 35.9% with GPT-4.1) and raises the valid response rate by over 6% compared to GPT-4.1 on the event-aware forecasting task.

cs.LG↗

IR$^3$: Contrastive Inverse Reinforcement Learning for Interpretable Detection and Mitigation of Reward Hacking

Reinforcement Learning from Human Feedback (RLHF) enables powerful LLM alignment but can introduce reward hacking - models exploit spurious correlations in proxy rewards without genuine alignment. Compounding this, the objectives internalized during RLHF remain opaque, making hacking behaviors difficult to detect or correct. We introduce IR3 (Interpretable Reward Reconstruction and Rectification), a framework that reverse-engineers, interprets, and surgically repairs the implicit objectives driving RLHF-tuned models. We propose Contrastive Inverse Reinforcement Learning (C-IRL), which reconstructs the implicit reward function by contrasting paired responses from post-alignment and baseline policies to explain behavioral shifts during RLHF. We then decompose the reconstructed reward via sparse autoencoders into interpretable features, enabling identification of hacking signatures through contribution analysis. Finally, we propose mitigation strategies - clean reward optimization, adversarial shaping, constrained optimization, and feature-guided distillation - that target problematic features while preserving beneficial alignment. Experiments across multiple reward model configurations show that IR3 achieves 0.89 correlation with ground-truth rewards, identifies hacking features with over 90% precision, and significantly reduces hacking behaviors while maintaining capabilities within 3% of the original model.

cs.AI↗

EIDOS: Latent-Space Predictive Learning for Time Series Foundation Models

Most time series foundation models are pretrained by directly predicting future observations, which often yields weakly structured latent representations that capture surface noise rather than coherent and predictable temporal dynamics. In this work, we introduce EIDOS, a foundation model family that shifts pretraining from future value prediction to latent-space predictive learning. We train a causal Transformer to predict the evolution of latent representations, encouraging the emergence of structured and temporally coherent latent states. To ensure stable targets for latent-space learning, we design a lightweight aggregation branch to construct target representations. EIDOS is optimized via a joint objective that integrates latent-space alignment, observational grounding to anchor representations to the input signal, and direct forecasting supervision. On the GIFT-Eval benchmark, EIDOS mitigates structural fragmentation in the representation space and achieves state-of-the-art performance. These results demonstrate that constraining models to learn predictable latent dynamics is a principled step toward more robust and reliable time series foundation models.

cs.LG↗