arXiv subjects

Zeyuan Li

Publications and source records attributed to Zeyuan Li.

18 recordsLinked to original sources

arXiv2026

Emergent Misaligned Communication in Long-Horizon Multi-Agent LLM Commerce

Frontier LLM agents increasingly transact on behalf of separate principals, often using natural language rather than structured APIs. Much of the safety literature studies misaligned LLM behavior through adversarial-elicitation evaluations on single agents or stylized tasks. Its prevalence and structure in settings that combine long horizons, separate principals, real operational state, and inter-agent natural-language exchange remain insufficiently measured. We study 2,583 inter-agent emails from 20 one-year simulation runs of Vending-Bench Arena, a competitive vending environment spanning 13 frontier LLMs. We operationalize speech-act misalignment as emails containing false factual claims, manipulation, collusion, or threats, combining message content with ground-truth simulator state and logged reasoning traces to classify and validate such behavior. Under our primary classifier, 12.6% of emails are labeled misaligned; misalignment appears in all 20 runs and 74.7% of individual agent-runs. Both the magnitude and composition of this misalignment are preserved under repeated classification at different sampling temperatures and under full-pipeline replication with judges from two other frontier-model families. Misalignment is also reciprocal and stress-conditioned: receiving a misaligned email from a counterparty raises the odds of a misaligned reply by 1.65x, and low-inventory conditions raise them by 1.58x. Across tests of capability-asymmetric exploitation, we find no evidence that higher-capability models differentially exploit weaker counterparties, and model performance rank does not predict misalignment rates. Together, these results indicate that measurable, state-dependent misalignment can arise in competitive multi-agent environments without engineered elicitation, in patterns associated with operational scarcity and counterparty behavior rather than model capability alone.

cs.MA↗

arXiv2026

Rethinking EEG-Based Disease Diagnosis: Decoupling Instance Representation Learning from Subject-Level Supervision

EEG-based disease diagnosis requires one prediction per subject, yet common pipelines segment recordings into short instances, inherit the subject label for every instance, and train instance-level classifiers. This assumes that all instances provide equally reliable diagnostic evidence. Multiple instance learning (MIL) avoids inherited labels by treating each subject as a bag. However, EEG datasets contain far fewer subjects than instances, which can limit the quality of the representations learned by end-to-end MIL. We propose BridgeMIL, a two-stage framework that decouples instance representation learning from subject-level supervision. Stage 1 pretrains the encoder without inherited instance labels by aligning temporally nearby windows and independently sampled within-subject sub-bags. Variance and covariance regularization prevent collapse and reduce redundancy without negative pairs. Stage 2 transfers the encoder to an attention-based MIL aggregator, applies supervision only to subject predictions, and limits representation drift through feature retention. Across three EEG disease datasets and five representative backbones, BridgeMIL attains the highest mean accuracy in 14 of 15 dataset-backbone settings and an overall mean accuracy of 76.57%, 4.28 percentage points higher than the strongest baseline. Further analyses reveal substantial variation in inherited-label reliability across instances, greater performance sensitivity to subject scarcity than to instance scarcity, and a more structured representation space with distinct subject-wise clusters and improved separation between diagnostic classes. Together, these findings underscore the importance of aligning supervision with the subject-level prediction objective while learning from abundant EEG instances without assigning disease labels to individual instances.

cs.LG↗

arXiv2026

Mirror Illusion Art

Mirror Illusion Art is a novel reflection-conditioned 3D illusion where one object yields two target appearances (front and mirror). The task is formulated as inverse design from two target 2D images (front and mirror) to a printable 3D object with geometry and texture. Prior topology-driven and shadow-based approaches demand substantial manual effort, optimize shape only, and often yield non-smooth or incomplete geometry. To address these challenges, we propose AutoMIA, an automated Mirror Illusion Art design pipeline that jointly optimizes shape and color. To stabilize optimization and suppress artifacts, four mechanisms are introduced: (1) projection-alignment component (PAC) selection to reduce surface noise, (2) position-weighted adaptive (PWA) suppression for background noise, (3) internal voxel preservation (IVP) to prevent internal fractures, and (4) shape-color decoupled (SCD) optimization that balance shape and color optimization. AutoMIA generate diverse smooth Mirror Illusion artworks successfully both in the digital and physical world, with only around 76s design time and 2.6 GB memory on average using a single RTX 3090, advancing inverse graphics and computational design. Our code is available at https://github.com/zxp555/AutoMIA.

cs.CV↗

arXiv2026

DSAINet: An Efficient Dual-Scale Attentive Interaction Network for General EEG Decoding

In real-world applications of noninvasive electroencephalography (EEG), specialized decoders often show limited generalizability across diverse tasks under subject-independent settings. One central challenge is that task-relevant EEG signals often follow different temporal organization patterns across tasks, while many existing methods rely on task-tailored architectural designs that introduce task-specific temporal inductive biases. This mismatch makes it difficult to adapt temporal modeling across tasks without changing the model configuration. To address these challenges, we propose DSAINet, an efficient dual-scale attentive interaction network for general EEG decoding. Specifically, DSAINet constructs shared spatiotemporal token representations from raw EEG signals and models diverse temporal dynamics through parallel convolutional branches at fine and coarse scales. The resulting representations are then adaptively refined by intra-branch attention to emphasize salient scale-specific patterns and by inter-branch attention to integrate task-relevant features across scales, followed by adaptive token aggregation to yield a compact representation for prediction. Extensive experiments on five downstream EEG decoding tasks across ten public datasets show that DSAINet consistently outperforms 13 representative baselines under strict subject-independent evaluation. Notably, this performance is achieved using the same architecture hyperparameters across datasets. Moreover, DSAINet achieves a favorable accuracy-efficiency trade-off with only about 77K trainable parameters and provides interpretable neurophysiological insights. The code is publicly available at https://github.com/zy0929/DSAINet.

cs.AI↗

arXiv2026

Bit-Efficient Quantisation for Two-Channel Modulo-Sampling Systems

Two-channel modulo analog-to-digital converters (ADCs) enable high-dynamic-range signal sensing at the Nyquist rate per channel, but existing designs quantise both channel outputs independently, incurring redundant bitrate costs. This paper proposes a bit-efficient quantisation scheme that exploits the integer-valued structure of inter-channel differences, transmitting one quantised channel output together with a compact difference index. We prove that this approach requires only 1-2 bits per signal sample overhead relative to conventional ADCs, despite operating with a much smaller per-channel dynamic range. Simulations confirm the theoretical error bounds and bitrate analysis, while hardware experiments demonstrate substantial bitrate savings compared with existing modulo sampling schemes, while maintaining comparable reconstruction accuracy. These results highlight a practical path towards high-resolution, bandwidth-efficient modulo ADCs for bitrate-constrained systems.

eess.SP↗

arXiv2026

Anisotropic Kinetics of Ion-Irradiation-Induced Phase Transition in Gallium Oxide

Radiation-tolerant semiconductors have traditionally been engineered by the principle of suppressing defect accumulation and amorphization, based on the assumption that radiation damage is inherently stochastic. Here we show that, in monoclinic $\beta$-\ce{Ga2O3}, a promising ultrawide-bandgap semiconductor, surface crystallographic orientation deterministically governs radiation tolerance through highly anisotropic kinetics of the $\beta$-to-$\gamma$ phase transition. Using machine-learning molecular dynamics coupled with a local configurational-entropy descriptor, we quantitatively map anisotropic $\beta$-to-$\gamma$ transition kinetics, showing that the critical dose, transition-layer depth, and kinetic stability of the $\gamma$-phase are fundamentally governed by surface orientation. Under ion irradiation, non-channeling surfaces such as (100), (001), and (-201) undergo severe surface amorphization, whereas the strongly channeling (010) surface resists damage accumulation and promotes subsurface $\gamma$-phase nucleation. During thermal annealing recovery process, these initial states follow two distinct recovery pathways: the channeling (010) surface reverts directly from $\gamma$-to-$\beta$, whereas non-channeling surfaces follow a sequential amorphous-to-$\gamma$-to-$\beta$ transition pathway. This work establishes surface orientation as a fundamental design principle for achieving radiation tolerance through controlled polymorphic transitions, providing a universal framework for engineering functional materials capable of withstanding extreme irradiation environments.

cond-mat.mtrl-sci↗

arXiv2025

FPGA-Enabled Modulo ADC with x100 Dynamic-Range Expansion: Hardware Design and Performance Evaluation

Conventional analog-to-digital converters (ADCs) fail to capture high-dynamic-range (HDR) signals due to clipping. Modulo ADCs circumvent this limitation by folding the input prior to quantization and algorithmically reconstructing the original waveform. This work presents a field-programmable gate array (FPGA)-based modulo ADC platform for systematic HDR performance evaluation. The mixed-signal architecture integrates a precision analog front end with a 200-MHz FPGA control loop that incorporates multi-bit updates and digital under-compensation calibration, ensuring stable folding and accurate feedback generation. The system achieves more than a hundred-fold dynamic-range expansion within a 400-kHz bandwidth while maintaining fidelity comparable to that of a conventional ADC. A system-on-chip (SoC)-like implementation enables on-board real-time recovery and supports benchmarking of state-of-the-art reconstruction algorithms, providing a compact and practical framework for HDR signal acquisition and evaluation.

eess.SP↗

arXiv2025

Difference-Based Recovery for Modulo Sampling: Tightened Bounds and Robustness Guarantees

Conventional analog-to-digital converters (ADCs) clip when signals exceed their input range. Modulo (unlimited) sampling overcomes this limitation by folding the signal before digitization, but existing recovery methods are either computationally intensive or constrained by loose oversampling bounds that demand high sampling rates. In addition, none account for sampling jitter, which is unavoidable in practice. This paper revisits difference-based recovery and establishes new theoretical and practical guarantees. In the noiseless setting, we prove that arbitrarily high difference order reduces the sufficient oversampling factor from $2\pi e$ to $\pi$, substantially tightening classical bounds. For fixed order $N$, we derive a noise-aware sampling condition that guarantees stable recovery. For second-order difference-based recovery ($N=2$), we further extend the analysis to non-uniform sampling, proving robustness under bounded jitter. An FPGA-based hardware prototype demonstrates reliable reconstruction with amplitude expansion up to $\rho = 108$, confirming the feasibility of high-performance unlimited sensing with a simple and robust recovery pipeline.

eess.SP↗

arXiv2025

Non-Foster Photonic Time Crystals

Photonic time crystals (PhTCs) are spatially uniform media whose material parameters vary periodically in time, opening momentum bandgaps within which the fields of electromagnetic modes can grow exponentially in time. To date, PhTCs have utilized only passive, lossless materials with "positive" dispersion (Foster materials), and a theoretical framework addressing active materials with "negative" dispersion (non-Foster materials) in PhTCs and their associated physical properties remains undeveloped. Here, we explore the two classes of isotropic PhTCs with embedded non-Foster inclusions: a bulk medium with periodically modulated negative permittivity, and a metasurface whose surface capacitance alternates between positive and negative values. Employing an analytical transfer-matrix formulation, we demonstrate that non-Foster permittivity modulation not only broadens momentum bandgaps without bounds but also provides a gain rate that increases linearly with momentum. Remarkably, the proposed isotropic PhTCs support exponential amplification down to zero frequency-a regime inaccessible in conventional isotropic PhTCs. These results open new avenues for ultra-broadband wave control, high-gain signal processing, and energy-harvesting devices that leverage the unique dispersion of active, time-modulated circuitry.

physics.optics↗

arXiv2025

eSapiens: A Platform for Secure and Auditable Retrieval-Augmented Generation

We present eSapiens, an AI-as-a-Service (AIaaS) platform engineered around a business-oriented trifecta: proprietary data, operational workflows, and any major agnostic Large Language Model (LLM). eSapiens gives businesses full control over their AI assets, keeping everything in-house for AI knowledge retention and data security. eSapiens AI Agents (Sapiens) empower your team by providing valuable insights and automating repetitive tasks, enabling them to focus on high-impact work and drive better business outcomes. The system integrates structured document ingestion, hybrid vector retrieval, and no-code orchestration via LangChain, and supports top LLMs including OpenAI, Claude, Gemini, and DeepSeek. A key component is the THOR Agent, which handles structured SQL-style queries and generates actionable insights over enterprise databases. To evaluate the system, we conduct two experiments. First, a retrieval benchmark on legal corpora reveals that a chunk size of 512 tokens yields the highest retrieval precision (Top-3 accuracy: 91.3%). Second, a generation quality test using TRACe metrics across five LLMs shows that eSapiens delivers more context-consistent outputs with up to a 23% improvement in factual alignment. These results demonstrate the effectiveness of eSapiens in enabling trustworthy, auditable AI workflows for high-stakes domains like legal and finance.

cs.AI↗

arXiv2025

THOR: Transformer Heuristics for On-Demand Retrieval

We introduce the THOR (Transformer Heuristics for On-Demand Retrieval) Module, designed and implemented by eSapiens, a secure, scalable engine that transforms natural-language questions into verified, read-only SQL analytics for enterprise databases. The Text-to-SQL module follows a decoupled orchestration/execution architecture: a Supervisor Agent routes queries, Schema Retrieval dynamically injects table and column metadata, and a SQL Generation Agent emits single-statement SELECT queries protected by a read-only guardrail. An integrated Self-Correction & Rating loop captures empty results, execution errors, or low-quality outputs and triggers up to five LLM-driven regeneration attempts. Finally, a Result Interpretation Agent produces concise, human-readable insights and hands raw rows to the Insight & Intelligence engine for visualization or forecasting. Smoke tests across finance, sales, and operations scenarios demonstrate reliable ad-hoc querying and automated periodic reporting. By embedding schema awareness, fault-tolerant execution, and compliance guardrails, the THOR Module empowers non-technical users to access live data with zero-SQL simplicity and enterprise-grade safety.

cs.DB↗

arXiv2025

eSapiens's DEREK Module: Deep Extraction & Reasoning Engine for Knowledge with LLMs

We present the DEREK (Deep Extraction & Reasoning Engine for Knowledge) Module, a secure and scalable Retrieval-Augmented Generation pipeline designed specifically for enterprise document question answering. Designed and implemented by eSapiens, the system ingests heterogeneous content (PDF, Office, web), splits it into 1,000-token overlapping chunks, and indexes them in a hybrid HNSW+BM25 store. User queries are refined by GPT-4o, retrieved via combined vector+BM25 search, reranked with Cohere, and answered by an LLM using CO-STAR prompt engineering. A LangGraph verifier enforces citation overlap, regenerating answers until every claim is grounded. On four LegalBench subsets, 1000-token chunks improve Recall@50 by approximately 1 pp and hybrid+rerank boosts Precision@10 by approximately 7 pp; the verifier raises TRACe Utilization above 0.50 and limits unsupported statements to less than 3%. All components run in containers, enforce end-to-end TLS 1.3 and AES-256. These results demonstrate that the DEREK module delivers accurate, traceable, and production-ready document QA with minimal operational overhead. The module is designed to meet enterprise demands for secure, auditable, and context-faithful retrieval, providing a reliable baseline for high-stakes domains such as legal and finance.

cs.CL↗

arXiv2025

eSapiens: A Real-World NLP Framework for Multimodal Document Understanding and Enterprise Knowledge Processing

We introduce eSapiens, a unified question-answering system designed for enterprise settings, which bridges structured databases and unstructured textual corpora via a dual-module architecture. The system combines a Text-to-SQL planner with a hybrid Retrieval-Augmented Generation (RAG) pipeline, enabling natural language access to both relational data and free-form documents. To enhance answer faithfulness, the RAG module integrates dense and sparse retrieval, commercial reranking, and a citation verification loop that ensures grounding consistency. We evaluate eSapiens on the RAGTruth benchmark across five leading large language models (LLMs), analyzing performance across key dimensions such as completeness, hallucination, and context utilization. Results demonstrate that eSapiens outperforms a FAISS baseline in contextual relevance and generation quality, with optional strict-grounding controls for high-stakes scenarios. This work provides a deployable framework for robust, citation-aware question answering in real-world enterprise applications.

cs.IR↗

arXiv2025

Option Pricing Using Ensemble Learning

Ensemble learning is characterized by flexibility, high precision, and refined structure. As a critical component within computational finance, option pricing with machine learning requires both high predictive accuracy and reduced structural complexity-features that align well with the inherent advantages of ensemble learning. This paper investigates the application of ensemble learning to option pricing, and conducts a comparative analysis with classical machine learning models to assess their performance in terms of accuracy, local feature extraction, and robustness to noise. A novel experimental strategy is introduced, leveraging parameter transfer across experiments to improve robustness and realism in financial simulations.Building upon this strategy, an evaluation mechanism is developed that incorporates a scoring strategy and a weighted evaluation strategy explicitly emphasizing the foundational role of financial theory. This mechanism embodies an orderly integration of theoretical finance and computational methods. In addition, the study examines the interaction between sliding window technique and noise, revealing nuanced patterns that suggest a potential connection relevant to ongoing research in machine learning and data science.

cs.LG↗

arXiv2024

Option Pricing with Convolutional Kolmogorov-Arnold Networks

With the rapid advancement of neural networks, methods for option pricing have evolved significantly. This study employs the Black-Scholes-Merton (B-S-M) model, incorporating an additional variable to improve the accuracy of predictions compared to the traditional Black-Scholes (B-S) model. Furthermore, Convolutional Kolmogorov-Arnold Networks (Conv-KANs) and Kolmogorov-Arnold Networks (KANs) are introduced to demonstrate that networks with enhanced non-linear capabilities yield superior fitting performance. For comparative analysis, Conv-LSTM and LSTM models, which are widely used in time series forecasting, are also applied. Additionally, a novel data selection strategy is proposed to simulate a real trading environment, thereby enhancing the robustness of the model.

cs.CE↗

arXiv2024

Volume-Preserving Transformers for Learning Time Series Data with Structure

Two of the many trends in neural network research of the past few years have been (i) the learning of dynamical systems, especially with recurrent neural networks such as long short-term memory networks (LSTMs) and (ii) the introduction of transformer neural networks for natural language processing (NLP) tasks. While some work has been performed on the intersection of these two trends, those efforts were largely limited to using the vanilla transformer directly without adjusting its architecture for the setting of a physical system. In this work we develop a transformer-inspired neural network and use it to learn a dynamical system. We (for the first time) change the activation function of the attention layer to imbue the transformer with structure-preserving properties to improve long-term stability. This is shown to be of great advantage when applying the neural network to learning the trajectory of a rigid body.

math.NA↗

arXiv2024

FALCON: Feedback-driven Adaptive Long/short-term memory reinforced Coding Optimization system

Recently, large language models (LLMs) have achieved significant progress in automated code generation. Despite their strong instruction-following capabilities, these models frequently struggled to align with user intent in coding scenarios. In particular, they were hampered by datasets that lacked diversity and failed to address specialized tasks or edge cases. Furthermore, challenges in supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF) led to failures in generating precise, human-intent-aligned code. To tackle these challenges and improve the code generation performance for automated programming systems, we propose Feedback-driven Adaptive Long/short-term memory reinforced Coding Optimization (i.e., FALCON). FALCON is structured into two hierarchical levels. From the global level, long-term memory improves code quality by retaining and applying learned knowledge. At the local level, short-term memory allows for the incorporation of immediate feedback from compilers and AI systems. Additionally, we introduce meta-reinforcement learning with feedback rewards to solve the global-local bi-level optimization problem and enhance the model's adaptability across diverse code generation tasks. Extensive experiments demonstrate that our technique achieves state-of-the-art performance, leading other reinforcement learning methods by more than 4.5 percentage points on the MBPP benchmark and 6.1 percentage points on the Humaneval benchmark. The open-sourced code is publicly available at https://github.com/titurte/FALCON.

cs.LG↗

arXiv2024

Orientation-dependent surface radiation damage in $β$-Ga2O3 explored by multiscale atomic simulations

Ultrawide bandgap semiconductor $β$-Ga2O3 holds extensive potential for applications in high-radiation environments. One of the primary challenges in its practical application is unveiling the mechanisms of surface irradiation damage under extreme conditions. In this study, we investigate the orientation-dependent mechanisms of radiation damage on four experimentally relevant $β$-Ga2O3 surface facets, namely, (100), (010), (001), and (-201), at various temperatures. We employ a multiscale atomic simulation approach, combining machine-learning-driven molecular dynamics (ML-MD) simulations and density functional theory (DFT) calculations. The results reveal that Ga vacancies and O interstitials are the predominant defects across all four surfaces, with the formation of many antisite defects Ga_O and few O_Ga observed. Among the two Ga sites and three O sites, the vacancy found in the O2 site is dominant, while the interstitials at the Ga1 and O1 sites are more significant. Interestingly, the (010) surface exhibits the lowest defect density, owing to its more profound channeling effect leading to a broader spread of defects. The influence of temperature on surface irradiation damage of $β$-Ga2O3 should be evaluated based on the unique crystal surface characteristics. Moreover, the formation energy and defect concentration calculated by DFT corroborate the results of the MD simulations. Comprehending surface radiation damage at the atomic level is crucial for assessing the radiation tolerance and predicting the performance changes of $β$-Ga2O3-based device in high-radiation environments.

cond-mat.mtrl-sci↗