arXiv subjects

Hengzhi Li

Publications and source records attributed to Hengzhi Li.

7 recordsLinked to original sources

arXiv2026

PuzzleWorld: A Benchmark for Multimodal, Open-Ended Reasoning in Puzzlehunts

Puzzlehunts are a genre of complex, multi-step puzzles lacking well-defined problem definitions. In contrast to conventional reasoning benchmarks consisting of tasks with clear instructions and constrained environments, puzzlehunts requires discovering the underlying problem structure from multimodal evidence and iterative reasoning, mirroring real-world domains such as scientific discovery, exploratory data analysis, or investigative problem-solving. Despite progress in foundation models, their performance on open-ended settings remains largely untested. We introduce PuzzleWorld, a comprehensive benchmark of 667 puzzlehunt-style problems designed to assess step-by-step, open-ended, and creative multimodal reasoning. Each puzzle is annotated with the final solution, detailed reasoning traces, and cognitive skill labels, enabling holistic benchmarking and fine-grained diagnostic analysis. Most state-of-the-art models achieve only 1-4% final answer accuracy. On PuzzleWorld, the best model solves only 18% of puzzles and reaches 40% stepwise accuracy, matching human puzzle novices but falling significantly behind puzzle enthusiasts. To demonstrate the value of our reasoning annotations, we show that fine-tuning a small model on reasoning traces boosts stepwise accuracy from 4% to 11%, which translates to improvements in downstream visual reasoning tasks. Our detailed error analysis reveals that current models exhibit myopic reasoning, are bottlenecked by the limitations of language-based inference, and lack sketching capabilities crucial for visual and spatial reasoning. We release PuzzleWorld at https://github.com/MIT-MI/PuzzleWorld to support future work on building more general, open-ended, and creative reasoning systems.

cs.CL↗

arXiv2026

Strong optical nonreciprocity in a photonic crystal composed of spinning cylinders

Moving media break time-reversal symmetry and exhibit intriguing optical nonreciprocity. This nonreciprocity is usually weak due to the much lower moving speed of media relative to the speed of light. We demonstrate that strong optical nonreciprocity can emerge in a two-dimensional photonic crystal composed of spinning dielectric cylinders. The photonic crystal supports two types of chiral modes at the Brillouin zone center: hybridized multipole modes and symmetry-protected bound states in the continuum (BICs), both of which carry intrinsic spin angular momentum. For finite wavevectors near the zone center, the BICs transform into quasi-bound states in the continuum (QBICs). Under oblique incidence of circularly polarized plane waves, the photonic crystal exhibits nonreciprocal transmission and absorption that are significantly enhanced at the frequencies of these hybridized multipole modes and QBICs. Furthermore, the high quality factors of the QBICs enable sharp transitions in nonreciprocity. Our work uncovers strong chiral light-matter interactions in periodic moving structures, with potential applications in nonreciprocal light manipulation. The mechanism may also be generalized to other classical wave systems, such as phononic crystals.

physics.optics↗

arXiv2026

Human Behavior Atlas: Benchmarking Unified Psychological and Social Behavior Understanding

Using intelligent systems to perceive psychological and social behaviors, that is, the underlying affective, cognitive, and pathological states that are manifested through observable behaviors and social interactions, remains a challenge due to their complex, multifaceted, and personalized nature. Existing work tackling these dimensions through specialized datasets and single-task systems often miss opportunities for scalability, cross-task transfer, and broader generalization. To address this gap, we curate Human Behavior Atlas, a unified benchmark of diverse behavioral tasks designed to support the development of foundation models for understanding psychological and social behaviors. Human Behavior Atlas comprises over 100,000 samples spanning text, audio, and visual modalities, covering tasks on affective states, cognitive states, pathologies, and social processes. Our unification efforts can reduce redundancy and cost, enable training to scale efficiently across tasks, and enhance generalization of behavioral features across domains. On Human Behavior Atlas, we train three models: Omnisapiens-7B SFT, Omnisapiens-7B BAM, and Omnisapiens-7B RL. We show that training on Human Behavior Atlas enables models to consistently outperform existing multimodal LLMs across diverse behavioral tasks. Pretraining on Human Behavior Atlas also improves transfer to novel behavioral datasets; with the targeted use of behavioral descriptors yielding meaningful performance gains. The benchmark, models, and codes can be found at: https://github.com/MIT-MI/human_behavior_atlas.

cs.AI↗

arXiv2025

MimeQA: Towards Socially-Intelligent Nonverbal Foundation Models

As AI becomes more closely integrated with peoples' daily activities, socially intelligent AI that can understand and interact seamlessly with humans in daily lives is increasingly important. However, current works in AI social reasoning all rely on language-only or language-dominant approaches to benchmark and training models, resulting in systems that are improving in verbal communication but struggle with nonverbal social understanding. To address this limitation, we tap into a novel data source rich in nonverbal social interactions -- mime videos. Mimes refer to the art of expression through gesture and movement without spoken words, which presents unique challenges and opportunities in interpreting nonverbal social communication. We contribute a new dataset called MimeQA, obtained by sourcing ~8 hours of videos clips from YouTube and developing a comprehensive video question-answering benchmark comprising 806 carefully annotated and verified question-answer pairs, designed to probe nonverbal social reasoning capabilities. Using MimeQA, we evaluate state-of-the-art video large language models (VideoLLMs) and find that they achieve low accuracy, generally ranging from 20-30%, while humans score 86%. Our analysis reveals that VideoLLMs often fail to ground imagined objects and over-rely on the text prompt while ignoring subtle nonverbal interactions. We hope to inspire future work in AI models that embody true social intelligence capable of interpreting non-verbal human interactions.

cs.CL↗

arXiv2025

Optical force and torque on a spinning dielectric sphere

Optical force can enable precise manipulations of small particles for various applications. It is well known that an isotropic lossless dielectric sphere is only subject to forward optical force under the illumination of an electromagnetic plane wave. By using rigorous full-wave simulations, we show that such a sphere can experience a lateral optical force and an optical torque besides the conventional longitudinal force, if it spins with a constant angular velocity. The emergence of the unusual optical force and torque is attributed to the breaking of mirror and cylindrical symmetries by the spinning motion. Using the multipole expansion in source representation, we illustrate how the spinning-induced effective bi-anisotropy generates the lateral force and torque on the sphere through the interference of electric and magnetic multipoles. We also uncover the effect of Sagnac frequency splitting on the optical force and torque. The results contribute to the understanding of the optical force and torque in moving media and can be applied to realize unconventional optical manipulations of small particles.

physics.optics↗

arXiv2024

Nonreciprocal optical metasurface based on spinning cylinders

Optical systems breaking Lorentz reciprocity have attracted broad attention due to their intriguing physics and applications. Nonreciprocal metasurfaces can enable one-way light transmission and reflection with essential applications in optical communication. Conventional nonreciprocal metasurfaces rely on using magneto-optic or nonlinear materials to induce nonreciprocal optical properties. Here, we propose and demonstrate a new mechanism for realizing nonreciprocal metasurfaces based on the relativistic effect of a moving medium. The metasurface is composed of periodic spinning dielectric cylinders located above a dielectric substrate. The spinning motion breaks the time-reversal symmetry and induces bi-anisotropic Tellegen-type response of the meta-atoms. We show that the metasurface can realize both asymmetric and nonreciprocal manipulations of the incident plane wave. The underlying mechanism is attributed to the Sagnac effect associated with the chiral multipole modes of the coupled spinning cylinders. By introducing dielectric pillars to modulate the phase profile, the metasurface can enable nonreciprocal wavefront manipulations. Our work offers a new mechanism for realizing nonreciprocal light manipulation in free space. The proposed metasurface can serve as a platform to explore the interesting physics of nonreciprocal optics, non-Hermitian optics, and topological photonics.

physics.optics↗

arXiv2023

Interactive Explanations by Conflict Resolution via Argumentative Exchanges

As the field of explainable AI (XAI) is maturing, calls for interactive explanations for (the outputs of) AI models are growing, but the state-of-the-art predominantly focuses on static explanations. In this paper, we focus instead on interactive explanations framed as conflict resolution between agents (i.e. AI models and/or humans) by leveraging on computational argumentation. Specifically, we define Argumentative eXchanges (AXs) for dynamically sharing, in multi-agent systems, information harboured in individual agents' quantitative bipolar argumentation frameworks towards resolving conflicts amongst the agents. We then deploy AXs in the XAI setting in which a machine and a human interact about the machine's predictions. We identify and assess several theoretical properties characterising AXs that are suitable for XAI. Finally, we instantiate AXs for XAI by defining various agent behaviours, e.g. capturing counterfactual patterns of reasoning in machines and highlighting the effects of cognitive biases in humans. We show experimentally (in a simulated environment) the comparative advantages of these behaviours in terms of conflict resolution, and show that the strongest argument may not always be the most effective.

cs.AI↗