arXiv subjects

Kevin Zhu

Publications and source records attributed to Kevin Zhu.

At least 73 records · Page 4Linked to original sources

arXiv2025

Grounding Foundational Vision Models with 3D Human Poses for Robust Action Recognition

For embodied agents to effectively understand and interact within the world around them, they require a nuanced comprehension of human actions grounded in physical space. Current action recognition models, often relying on RGB video, learn superficial correlations between patterns and action labels, so they struggle to capture underlying physical interaction dynamics and human poses in complex scenes. We propose a model architecture that grounds action recognition in physical space by fusing two powerful, complementary representations: V-JEPA 2's contextual, predictive world dynamics and CoMotion's explicit, occlusion-tolerant human pose data. Our model is validated on both the InHARD and UCF-19-Y-OCC benchmarks for general action recognition and high-occlusion action recognition, respectively. Our model outperforms three other baselines, especially within complex, occlusive scenes. Our findings emphasize a need for action recognition to be supported by spatial understanding instead of statistical pattern recognition.

cs.CV↗

arXiv2025

Inference-Time Chain-of-Thought Pruning with Latent Informativeness Signals

Large language models (LLMs) improve reasoning accuracy when generating multiple candidate solutions at test time, but standard methods like Best-of-N (BoN) incur high computational cost by fully generating all branches. Self-Truncation Best-of-N (ST-BoN) mitigates this by truncating unpromising paths early, but its reliance on consistency-based heuristics is a limitation as it does not directly evaluate branch quality. We present KL-Adjusted Pruned Path Algorithm (KAPPA), an inference-time method that combines Kullback-Leibler divergence, confidence, and entropy into a principled scoring function to guide progressive pruning. By promoting diversity during exploration and selectively eliminating low-scoring branches, KAPPA maintains accuracy while substantially reducing memory and token usage. Experiments on GSM8K and MATH500 with DeepSeek-R1-Distill-Qwen-1.5B and Qwen2.5-7B-Instruct demonstrate that KAPPA stabilizes performance in smaller models and achieves up to ~60% reduction in peak memory and ~90% reduction in total token generation relative to BoN, with minimal impact on accuracy.

cs.LG↗

arXiv2025

Shared Parameter Subspaces and Cross-Task Linearity in Emergently Misaligned Behavior

Recent work has discovered that large language models can develop broadly misaligned behaviors after being fine-tuned on narrowly harmful datasets, a phenomenon known as emergent misalignment (EM). However, the fundamental mechanisms enabling such harmful generalization across disparate domains remain poorly understood. In this work, we adopt a geometric perspective to study EM and demonstrate that it exhibits a fundamental cross-task linear structure in how harmful behavior is encoded across different datasets. Specifically, we find a strong convergence in EM parameters across tasks, with the fine-tuned weight updates showing relatively high cosine similarities, as well as shared lower-dimensional subspaces as measured by their principal angles and projection overlaps. Furthermore, we also show functional equivalence via linear mode connectivity, wherein interpolated models across narrow misalignment tasks maintain coherent, broadly misaligned behavior. Our results indicate that EM arises from different narrow tasks discovering the same set of shared parameter directions, suggesting that harmful behaviors may be organized into specific, predictable regions of the weight landscape. By revealing this fundamental connection between parametric geometry and behavioral outcomes, we hope our work catalyzes further research on parameter space interpretability and weight-based interventions.

cs.LG↗

arXiv2025

Interpreting the Latent Structure of Operator Precedence in Language Models

Large Language Models (LLMs) have demonstrated impressive reasoning capabilities but continue to struggle with arithmetic tasks. Prior works largely focus on outputs or prompting strategies, leaving the open question of the internal structure through which models do arithmetic computation. In this work, we investigate whether LLMs encode operator precedence in their internal representations via the open-source instruction-tuned LLaMA 3.2-3B model. We constructed a dataset of arithmetic expressions with three operands and two operators, varying the order and placement of parentheses. Using this dataset, we trace whether intermediate results appear in the residual stream of the instruction-tuned LLaMA 3.2-3B model. We apply interpretability techniques such as logit lens, linear classification probes, and UMAP geometric visualization. Our results show that intermediate computations are present in the residual stream, particularly after MLP blocks. We also find that the model linearly encodes precedence in each operator's embeddings post attention layer. We introduce partial embedding swap, a technique that modifies operator precedence by exchanging high-impact embedding dimensions between operators.

cs.CL↗

arXiv2025

SMAGDi: Socratic Multi Agent Interaction Graph Distillation for Efficient High Accuracy Reasoning

Multi-agent systems (MAS) often achieve higher reasoning accuracy than single models, but their reliance on repeated debates across agents makes them computationally expensive. We introduce SMAGDi, a distillation framework that transfers the debate dynamics of a five-agent Llama-based MAS into a compact Socratic decomposer-solver student. SMAGDi represents debate traces as directed interaction graphs, where nodes encode intermediate reasoning steps with correctness labels and edges capture continuity and cross-agent influence. The student is trained with a composite objective combining language modeling, graph-based supervision, contrastive reasoning, and embedding alignment to preserve both fluency and structured reasoning. On StrategyQA and MMLU, SMAGDi compresses a 40B multi-agent system into a 6B student while retaining 88% of its accuracy, substantially outperforming prior distillation methods such as MAGDi, standard KD, and fine-tuned baselines. These results highlight that explicitly modeling interaction graphs and Socratic decomposition enable small models to inherit the accuracy benefits of multi-agent debate while remaining efficient enough for real-world deployment.

cs.AI↗

arXiv2025

SwiftSolve: A Self-Iterative, Complexity-Aware Multi-Agent Framework for Competitive Programming

Correctness alone is insufficient: LLM-generated programs frequently satisfy unit tests while violating contest time or memory budgets. We present SwiftSolve, a complexity-aware multi-agent system for competitive programming that couples algorithmic planning with empirical profiling and complexity-guided repair. We frame competitive programming as a software environment where specialized agents act as programmers, each assuming roles such as planning, coding, profiling, and complexity analysis. A Planner proposes an algorithmic sketch; a deterministic Static Pruner filters high-risk plans; a Coder emits ISO C++17; a Profiler compiles and executes candidates on a fixed input-size schedule to record wall time and peak memory; and a Complexity Analyst fits log-log growth (s, R2) with an LLM fallback to assign a complexity class and dispatch targeted patches to either the Planner or Coder. Agents communicate via typed, versioned JSON; a controller enforces iteration caps and diminishing returns stopping. Evaluated on 26 problems (16 BigO, 10 Codeforces Div. 2) in a POSIX sandbox (2 s / 256-512 MB), SwiftSolve attains pass@1 = 61.54% (16/26) on the first attempt and Solved@<=3 = 80.77% with marginal latency change (mean 11.96 s to 12.66 s per attempt). Aggregate run-level success is 73.08% at 12.40 s mean. Failures are predominantly resource-bound, indicating inefficiency rather than logic errors. Against Claude Opus 4, SwiftSolve improves run-level success (73.1% vs 52.6%) at approximately 2x runtime overhead (12.4 s vs 6.8 s). Beyond correctness (pass@k), we report efficiency metrics (eff@k for runtime and memory, incidence of TLE or MLE, and complexity fit accuracy on BigO), demonstrating that profiling and complexity-guided replanning reduce inefficiency while preserving accuracy.

cs.AI↗

arXiv2025

Analytical Swarm Chemistry: Characterization and Analysis of Emergent Swarm Behaviors

Swarm robotics has potential for a wide variety of applications, but real-world deployments remain rare due to the difficulty of predicting emergent behaviors arising from simple local interactions. Traditional engineering approaches design controllers to achieve desired macroscopic outcomes under idealized conditions, while agent-based and artificial life studies explore emergent phenomena in a bottom-up, exploratory manner. In this work, we introduce Analytical Swarm Chemistry, a framework that integrates concepts from engineering, agent-based and artificial life research, and chemistry. This framework combines macrostate definitions with phase diagram analysis to systematically explore how swarm parameters influence emergent behavior. Inspired by concepts from chemistry, the framework treats parameters like thermodynamic variables, enabling visualization of regions in parameter space that give rise to specific behaviors. Applying this framework to agents with minimally viable capabilities, we identify sufficient conditions for behaviors such as milling and diffusion and uncover regions of the parameter space that reliably produce these behaviors. Preliminary validation on real robots demonstrates that these regions correspond to observable behaviors in practice. By providing a principled, interpretable approach, this framework lays the groundwork for predictable and reliable emergent behavior in real-world swarm systems.

cs.RO↗

arXiv2025

DuoLens: A Framework for Robust Detection of Machine-Generated Multilingual Text and Code

The prevalence of Large Language Models (LLMs) for generating multilingual text and source code has only increased the imperative for machine-generated content detectors to be accurate and efficient across domains. Current detectors, predominantly utilizing zero-shot methods, such as Fast DetectGPT or GPTZero, either incur high computational cost or lack sufficient accuracy, often with a trade-off between the two, leaving room for further improvement. To address these gaps, we propose the fine-tuning of encoder-only Small Language Models (SLMs), in particular, the pre-trained models of RoBERTA and CodeBERTa using specialized datasets on source code and other natural language to prove that for the task of binary classification, SLMs outperform LLMs by a huge margin whilst using a fraction of compute. Our encoders achieve AUROC $= 0.97$ to $0.99$ and macro-F1 $0.89$ to $0.94$ while reducing latency by $8$-$12\times$ and peak VRAM by $3$-$5\times$ at $512$-token inputs. Under cross-generator shifts and adversarial transformations (paraphrase, back-translation; code formatting/renaming), performance retains $\geq 92%$ of clean AUROC. We release training and evaluation scripts with seeds and configs; a reproducibility checklist is also included.

cs.CL↗

arXiv2025

AgentChangeBench: A Multi-Dimensional Evaluation Framework for Goal-Shift Robustness in Conversational AI

Goal changes are a defining feature of real world multi-turn interactions, yet current agent benchmarks primarily evaluate static objectives or one-shot tool use. We introduce AgentChangeBench, a benchmark explicitly designed to measure how tool augmented language model agents adapt to mid dialogue goal shifts across three enterprise domains. Our framework formalizes evaluation through four complementary metrics: Task Success Rate (TSR) for effectiveness, Tool Use Efficiency (TUE) for reliability, Tool Call Redundancy Rate (TCRR) for wasted effort, and Goal-Shift Recovery Time (GSRT) for adaptation latency. AgentChangeBench comprises 2,835 task sequences and five user personas, each designed to trigger realistic shift points in ongoing workflows. Using this setup, we evaluate several frontier models and uncover sharp contrasts obscured by traditional $\text{pass}@k$ scores: for example, GPT-4o reaches $92.2\%$ recovery on airline booking shifts while Gemini collapses to $48.6\%$, and retail tasks show near perfect parameter validity yet redundancy rates above $80\%$, revealing major inefficiencies. These findings demonstrate that high raw accuracy does not imply robustness under dynamic goals, and that explicit measurement of recovery time and redundancy is essential. AgentChangeBench establishes a reproducible testbed for diagnosing and improving agent resilience in realistic enterprise settings.

cs.AI↗

arXiv2025

AAVENUE: Detecting LLM Biases on NLU Tasks in AAVE via a Novel Benchmark

Detecting biases in natural language understanding (NLU) for African American Vernacular English (AAVE) is crucial to developing inclusive natural language processing (NLP) systems. To address dialect-induced performance discrepancies, we introduce AAVENUE ({AAVE} {N}atural Language {U}nderstanding {E}valuation), a benchmark for evaluating large language model (LLM) performance on NLU tasks in AAVE and Standard American English (SAE). AAVENUE builds upon and extends existing benchmarks like VALUE, replacing deterministic syntactic and morphological transformations with a more flexible methodology leveraging LLM-based translation with few-shot prompting, improving performance across our evaluation metrics when translating key tasks from the GLUE and SuperGLUE benchmarks. We compare AAVENUE and VALUE translations using five popular LLMs and a comprehensive set of metrics including fluency, BARTScore, quality, coherence, and understandability. Additionally, we recruit fluent AAVE speakers to validate our translations for authenticity. Our evaluations reveal that LLMs consistently perform better on SAE tasks than AAVE-translated versions, underscoring inherent biases and highlighting the need for more inclusive NLP models. We have open-sourced our source code on GitHub and created a website to showcase our work at https://aavenuee.github.io.

cs.CL↗

arXiv2025

Causal Language Control in Multilingual Transformers via Sparse Feature Steering

Deterministically controlling the target generation language of large multilingual language models (LLMs) remains a fundamental challenge, particularly in zero-shot settings where neither explicit language prompts nor fine-tuning are available. In this work, we investigate whether sparse autoencoder (SAE) features, previously shown to correlate with interpretable model behaviors, can be leveraged to steer the generated language of LLMs during inference. Leveraging pretrained SAEs on the residual streams of Gemma-2B and Gemma-9B, we identify features whose activations differ most significantly between English and four target languages: Chinese, Japanese, Spanish, and French. By modifying just a single SAE feature at one transformer layer, we achieve controlled language shifts with up to 90\% success, as measured by FastText language classification, while preserving semantic fidelity according to LaBSE (Language-Agnostic BERT Sentence Embedding) similarity. Our analysis reveals that language steering is most effective in mid-to-late transformer layers and is amplified by specific attention heads disproportionately associated with language-sensitive SAE features. These results demonstrate the promise of sparse feature steering as a lightweight and interpretable mechanism for controllable multilingual generation.

cs.CL↗

arXiv2025

ERGO: Entropy-guided Resetting for Generation Optimization in Multi-turn Language Models

Large Language Models (LLMs) suffer significant performance degradation in multi-turn conversations when information is presented incrementally. Given that multi-turn conversations characterize everyday interactions with LLMs, this degradation poses a severe challenge to real world usability. We hypothesize that abrupt increases in model uncertainty signal misalignment in multi-turn LLM interactions, and we exploit this insight to dynamically realign conversational context. We introduce ERGO (Entropy-guided Resetting for Generation Optimization), which continuously quantifies internal uncertainty via Shannon entropy over next token distributions and triggers adaptive prompt consolidation when a sharp spike in entropy is detected. By treating uncertainty as a first class signal rather than a nuisance to eliminate, ERGO embraces variability in language and modeling, representing and responding to uncertainty. In multi-turn tasks with incrementally revealed instructions, ERGO yields a 56.6% average performance gain over standard baselines, increases aptitude (peak performance capability) by 24.7%, and decreases unreliability (variability in performance) by 35.3%, demonstrating that uncertainty aware interventions can improve both accuracy and reliability in conversational AI.

cs.CL↗

arXiv2025

Limits of Emergent Reasoning of Large Language Models in Agentic Frameworks for Deterministic Games

Recent work reports that Large Reasoning Models (LRMs) undergo a collapse in performance on solving puzzles beyond certain perplexity thresholds. In subsequent discourse, questions have arisen as to whether the nature of the task muddles an evaluation of true reasoning. One potential confound is the requirement that the model keep track of the state space on its own. We provide a large language model (LLM) with an environment interface for Tower of Hanoi problems, allowing it to make a move with a tool call, provide written justification, observe the resulting state space, and reprompt itself for the next move. We observe that access to an environment interface does not delay or eradicate performance collapse. Furthermore, LLM-parameterized policy analysis reveals increasing divergence from both optimal policies and uniformly random policies, suggesting that the model exhibits mode-like collapse at each level of complexity, and that performance is dependent upon whether the mode reflects the correct solution for the problem. We suggest that a similar phenomena might take place in LRMs.

cs.AI↗

arXiv2025

Adaptive Originality Filtering: Rejection Based Prompting and RiddleScore for Culturally Grounded Multilingual Riddle Generation

Language models are increasingly tested on multilingual creativity, demanding culturally grounded, abstract generations. Standard prompting methods often produce repetitive or shallow outputs. We introduce Adaptive Originality Filtering (AOF), a prompting strategy that enforces novelty and cultural fidelity via semantic rejection. To assess quality, we propose RiddleScore, a metric combining novelty, diversity, fluency, and answer alignment. AOF improves Distinct-2 (0.915 in Japanese), reduces Self-BLEU (0.177), and raises RiddleScore (up to +57.1% in Arabic). Human evaluations confirm fluency, creativity, and cultural fit gains. However, improvements vary: Arabic shows greater RiddleScore gains than Distinct-2; Japanese sees similar changes. Though focused on riddles, our method may apply to broader creative tasks. Overall, semantic filtering with composite evaluation offers a lightweight path to culturally rich generation without fine-tuning.

cs.CL↗

arXiv2025

Evaluating the Impact of Radiographic Noise on Chest X-ray Semantic Segmentation and Disease Classification Using a Scalable Noise Injection Framework

Deep learning models are increasingly used for radiographic analysis, but their reliability is challenged by the stochastic noise inherent in clinical imaging. A systematic, cross-task understanding of how different noise types impact these models is lacking. Here, we evaluate the robustness of state-of-the-art convolutional neural networks (CNNs) to simulated quantum (Poisson) and electronic (Gaussian) noise in two key chest X-ray tasks: semantic segmentation and pulmonary disease classification. Using a novel, scalable noise injection framework, we applied controlled, clinically-motivated noise severities to common architectures (UNet, DeepLabV3, FPN; ResNet, DenseNet, EfficientNet) on public datasets (Landmark, ChestX-ray14). Our results reveal a stark dichotomy in task robustness. Semantic segmentation models proved highly vulnerable, with lung segmentation performance collapsing under severe electronic noise (Dice Similarity Coefficient drop of 0.843), signifying a near-total model failure. In contrast, classification tasks demonstrated greater overall resilience, but this robustness was not uniform. We discovered a differential vulnerability: certain tasks, such as distinguishing Pneumothorax from Atelectasis, failed catastrophically under quantum noise (AUROC drop of 0.355), while others were more susceptible to electronic noise. These findings demonstrate that while classification models possess a degree of inherent robustness, pixel-level segmentation tasks are far more brittle. The task- and noise-specific nature of model failure underscores the critical need for targeted validation and mitigation strategies before the safe clinical deployment of diagnostic AI.

eess.IV↗

arXiv2025

Adapting Large Language Models to Mitigate Skin Tone Biases in Clinical Dermatology Tasks: A Mixed-Methods Study

SkinGPT-4, a large vision-language model, leverages annotated skin disease images to augment clinical workflows in underserved communities. However, its training dataset predominantly represents lighter skin tones, limiting diagnostic accuracy for darker tones. Here, we evaluated performance biases in SkinGPT-4 across skin tones on common skin diseases, including eczema, allergic-contact dermatitis, and psoriasis using the open-sourced SCIN dataset. We leveraged the SkinGPT-4 backbone to develop finetuned models for custom skin disease classification tasks and explored bias mitigation strategies. Clinical evaluation by board-certified dermatologists on six relevant skin diseases from 300 SCIN cases assessed images for diagnostic accuracy, informativity, physician utility, and patient utility. Model fairness metrics, including demographic parity and equalized odds, were calculated across skin tones. SkinGPT-4 achieved an average demographic parity of 0.10 across Fitzpatrick types, with notable differences of 0.10-0.15 between lightest and darkest tones across evaluation metrics. Model hallucinations in artifacts and anatomy occurred at a rate of 17.8. Our customized models achieved average F1, precision, and AUROC of 0.75, 0.78, and 0.78 across visually similar disease pairs. Fairness analysis showed an average demographic parity of 0.75, with a maximum disparity of 0.21 across skin tones. The best model achieved parity scores of 0.83, 0.83, 0.76, 0.89, 0.90, and 0.90 for Fitzpatrick I-VI, indicating robust fairness. Large language models such as SkinGPT-4 showed weaker performance on darker tones. Model biases exist across evaluation criteria, and hallucinations may affect diagnostic efficacy. These findings demonstrate the efficacy of training accurate, fair models using existing backbones for custom skin disease classification.

eess.IV↗

arXiv2025

PALADIN: Self-Correcting Language Model Agents to Cure Tool-Failure Cases

Tool-augmented language agents frequently fail in real-world deployment due to tool malfunctions--timeouts, API exceptions, or inconsistent outputs--triggering cascading reasoning errors and task abandonment. Existing agent training pipelines optimize only for success trajectories, failing to expose models to the tool failures that dominate real-world usage. We propose \textbf{PALADIN}, a generalizable framework for equipping language agents with robust failure recovery capabilities. PALADIN trains on 50,000+ recovery-annotated trajectories constructed via systematic failure injection and expert demonstrations on an enhanced ToolBench dataset. Training uses LoRA-based fine-tuning to retain base capabilities while injecting recovery competence. At inference, PALADIN detects execution-time errors and retrieves the most similar case from a curated bank of 55+ failure exemplars aligned with ToolScan's taxonomy, then executes the corresponding recovery action. This approach generalizes to novel failures beyond the training distribution, retaining 95.2\% recovery performance on unseen tool APIs. Evaluation across PaladinEval and ToolReflectEval demonstrates consistent improvements in Recovery Rate (RR), Task Success Rate (TSR), Catastrophic Success Rate (CSR), and Efficiency Score (ES). PALADIN improves RR from 32.76% to 89.68% (+57% relative) over ToolBench and outperforms the strongest baseline CRITIC (76.34%) by +13.3%. Against vanilla agents, PALADIN achieves 89.86\% RR (+66% relative improvement from 23.75%). These results establish PALADIN as an effective method for building fault-tolerant agents capable of robust recovery in real-world tool environments.

cs.LG↗

arXiv2025

Reliable Weak-to-Strong Monitoring of LLM Agents

We stress test monitoring systems for detecting covert misbehavior in autonomous LLM agents (e.g., secretly sharing private information). To this end, we systematize a monitor red teaming (MRT) workflow that incorporates: (1) varying levels of agent and monitor situational awareness; (2) distinct adversarial strategies to evade the monitor, such as prompt injection; and (3) two datasets and environments -- SHADE-Arena for tool-calling agents and our new CUA-SHADE-Arena, which extends TheAgentCompany, for computer-use agents. We run MRT on existing LLM monitor scaffoldings, which orchestrate LLMs and parse agent trajectories, alongside a new hybrid hierarchical-sequential scaffolding proposed in this work. Our empirical results yield three key findings. First, agent awareness dominates monitor awareness: an agent's knowledge that it is being monitored substantially degrades the monitor's reliability. On the contrary, providing the monitor with more information about the agent is less helpful than expected. Second, monitor scaffolding matters more than monitor awareness: the hybrid scaffolding consistently outperforms baseline monitor scaffolding, and can enable weaker models to reliably monitor stronger agents -- a weak-to-strong scaling effect. Third, in a human-in-the-loop setting where humans discuss with the LLM monitor to get an updated judgment for the agent's behavior, targeted human oversight is most effective; escalating only pre-flagged cases to human reviewers improved the TPR by approximately 15% at FPR = 0.01. Our work establishes a standard workflow for MRT, highlighting the lack of adversarial robustness for LLMs and humans when monitoring and detecting agent misbehavior. We release code, data, and logs to spur further research.

cs.AI↗