Searcharxiv⌕ Search

arXiv subjects

Cong Chen

Publications and source records attributed to Cong Chen.

At least 37 records · Page 2Linked to original sources
arXiv2026

CAPE: Contrastive Action-conditioned Parallel Encoding for Embodied Planning

Embodied agents need to predict the future consequences of candidate actions in order to plan effectively before execution. Existing visual dynamics models learn by reconstructing future visual states or rolling out dense latent representations, which spreads learning capacity across visually salient but planning-irrelevant content rather than the action-conditioned changes that drive manipulation outcomes. We propose CAPE, a Contrastive Action-conditioned Parallel Encoding framework that learns visual dynamics by distinguishing the future outcomes induced by different action sequences. Given an initial observation and a candidate action sequence, CAPE decodes the full future latent trajectory in a single forward pass and is trained with a Goal-Convergent Contrastive Objective that aligns predictions corresponding to the same future outcome while separating those corresponding to different outcomes. On real-world DROID and zero-shot transfer to RoboCasa, CAPE substantially outperforms prior baselines on future-state retrieval, offline action matching, and closed-loop planning, while notably reducing planning-time inference cost at long prediction horizons.

cs.RO↗
arXiv2026

Eliciting Complex Spatial Reasoning in MLLMs through Wide-Baseline Matching

Wide-baseline matching (WBM) requires integrating geometric understanding, viewpoint changes, fine-grained perception, and occlusion reasoning, making it a challenging testbed for spatial reasoning in multimodal large language models (MLLMs) deployed in physical environments. However, current MLLMs lack systematic evaluation and training frameworks for these capabilities. We introduce ReasonMatch-Bench, a benchmark stratified by viewpoint displacement and matching granularity across indoor, outdoor, and object-centric scenarios, and show that current MLLMs still struggle with fine-grained wide-baseline correspondence: on a difficult 90-sample subset, human annotators achieve 84.0 F1, while the best existing baseline reaches 37.2. To bridge this gap, we build a scalable data-generation pipeline that automatically extracts wide-baseline view pairs from large-scale video-3D corpora, including RGB-D videos and SfM reconstructions, yielding diverse and verifiable supervision. We further propose Dynamic Correspondence Reinforcement Learning (DCRL), which combines Image-Level Viewpoint Progression and Point-Level Correspondence Curriculum to improve WBM training through verifiable rewards without explicit CoT supervision. Extensive experiments show that DCRL substantially improves ReasonMatch-Bench and transfers to related spatial benchmarks, while maintaining general visual understanding performance with modest gains on several benchmarks.

cs.CV↗
arXiv2026

A Scalable Bundle Method for Exact Reformulation of SDP in Three-Phase Power Flow Feasibility

Power flow feasibility assessment is computationally challenging for unbalanced three-phase distribution networks. This paper develops a vectorized semidefinite program (SDP) based on the bus injection model (BIM) and reformulates its dual as an exact-penalty problem, enabling us to develop a scalable three-cut proximal bundle method for feasibility assessment. The proposed bundle method is numerically over 400 times faster than MOSEK with less than 1/2000 of its memory; on the decomposed BIM-SDP, approximately 2 times faster with 75% less memory.

math.OC↗
arXiv2026

The Great Pretender: A Stochasticity Problem in LLM Jailbreak

"Oh-Oh, yes, I'm the great pretender. Pretending that I'm doing well. My need is such, I pretend too much..." summarizes the state in the area of jailbreak creation and evaluation. You find this method to generate adversarial attacks proposed by a reputable institution (e.g., BoN from Anthropic or Crescendo from Microsoft Research). However, this method does not deliver on the promise claimed in the paper despite having top ASR scores against industry-grade LLMs. You successfully generate the jailbreak prompts against your target (open) model. However, the generated jailbreak prompt works against the target model with a 50% consecutive success rate (5 out of 10 attempts) despite having an 80% ASR (on paper) on the latest closed-source model (with a guardrail system)! This observation leads us to think. First, Attack Success Rate (ASR), the primary metric for LLM jailbreak benchmarking, is not a stable quantity. Second, published ASR numbers are therefore systematically inflated and incomparable across papers. Therefore, we wonder "Why a successful jailbreak prompt does not perform consistently well against a target model on which the prompts have been optimized?". To answer this question, we study the impact of stochasticity not only during attack evaluation but also during attack generation. Our evaluation includes several jailbreak attacks, models (different sizes and providers), and judges. In addition, we propose a new metric and two new frameworks (CAS-eval and CAS-gen). Our evaluation framework, CAS-eval, shows that an attack can have an ASR drop of up to 30 percentage points when a jailbreak prompt needs to succeed on more than one attempt. Thankfully, our attack generation framework (CAS-gen) improves previous jailbreak methods and helps them recover this loss of 30 percentage points!

cs.CR↗
arXiv2026

Android Coach: Improve Online Agentic Training Efficiency with Single State Multiple Actions

Online reinforcement learning (RL) serves as an effective method for enhancing the capabilities of Android agents. However, guiding agents to learn through online interaction is prohibitively expensive due to the high latency of emulators and the sample inefficiency of existing RL algorithms. We identify a fundamental limitation in current approaches: the Single State Single Action paradigm, which updates the policy with one-to-one state-action pairs from online one-way rollouts without fully exploring each costly emulator state. In this paper, we propose Android Coach, a novel framework that shifts the training paradigm to Single State Multiple Actions, allowing the agent to sample and utilize multiple actions for a single online state. We enable this without additional emulator overhead by learning a critic that estimates action values. To ensure the critic serves as a reliable coach, we integrate a process reward model and introduce a group-wise advantage estimator based on the averaged critic outputs. Extensive experiments demonstrate the effectiveness and efficiency of Android Coach: it achieves 7.5% and 8.3% success rate improvements on AndroidLab and AndroidWorld over UI-TARS-1.5-7B, and attains 1.4x higher training efficiency than Single State Single Action methods PPO and GRPO at matched success rates.

cs.LG↗
arXiv2026

Mean-Field Learning for Storage Aggregation

Distributed energy storage devices can be aggregated to provide operational flexibility for power systems. This requires representing a massive device population as a single, tractable surrogate that is computationally efficient and accurate. However, surrogate identification is challenging due to heterogeneity, nonconvexity, and high dimensionality of storage devices. To address these challenges, this paper develops a mean-field learning framework for storage aggregation. We interpret aggregation as the average behavior of a large storage population and show that, as the population grows, aggregate performance converges to a unique, convex mean-field limit, enabling tractable population-level modeling. This convexity further yields a price-responsive characterization of aggregate storage behavior and allows us to bound the mean-field approximation error. We construct a convex surrogate model with physically interpretable parameters that approximates the aggregate behavior of large storage populations and can be embedded directly into power system operations. Surrogate parameter identification is formulated as an optimization problem using historical price-response data, and we adopt a gradient-based algorithm for efficient learning. Case studies validate the theoretical findings and demonstrate the effectiveness of the proposed framework in approximation accuracy and data efficiency.

eess.SY↗
arXiv2026

Wholesale Market Participation via Competitive DER Aggregation

We consider the aggregation of distributed energy resources (DERs), such as solar PV, energy storage, and flexible loads, by a profit-seeking aggregator participating directly in the wholesale market under distribution network access constraints. We propose a competitive DER aggregator (DERA) model that directly controls local DERs to maximize its profits, while ensuring each aggregated customer gains a surplus higher than their surplus under the regulated retail tariff. The DERA participates in the wholesale electricity market as virtual storage with optimized generation offers and consumption bids derived from the propoed competitive aggregation model. Also derived are DERA's bid curves for the distribution network access and DERA's profitability when competing with the regulated retail tariff. We show that, with the same distribution network access, the proposed DERA's wholesale market participation achieves the same welfare-maximizing outcome as when its customers participate directly in the wholesale market. Extensive numerical studies compare the proposed DERA with existing methods in terms of customer surplus and DERA profit. We empirically evaluate how many DERAs can survive in the competition at long-run equilibrium, and assess the impacts of DER adoption levels and distribution network access on short-run operations.

eess.SY↗
arXiv2026

Competition and Cooperation of LLM Agents in Games

Large language model (LLM) agents are increasingly deployed in competitive multi-agent settings, raising fundamental questions about whether they converge to equilibria and how their strategic behavior can be characterized. In this paper, we study LLM agent interactions in two standard games: a network resource allocation game and a Cournot competition game. Rather than converging to Nash equilibria, we find that LLM agents tend to cooperate when given multi-round prompts and non-zero-sum context. Chain-of-thought analysis reveals that fairness reasoning is central to this behavior. We propose an analytical framework that captures the dynamics of LLM agent reasoning across rounds and explains these experimental findings.

cs.MA↗
arXiv2026

TC-AE: Unlocking Token Capacity for Deep Compression Autoencoders

We propose TC-AE, a ViT-based architecture for deep compression autoencoders. Existing methods commonly increase the channel number of latent representations to maintain reconstruction quality under high compression ratios. However, this strategy often leads to latent representation collapse, which degrades generative performance. Instead of relying on increasingly complex architectures or multi-stage training schemes, TC-AE addresses this challenge from the perspective of the token space, the key bridge between pixels and image latents, through two complementary innovations: Firstly, we study token number scaling by adjusting the patch size in ViT under a fixed latent budget, and identify aggressive token-to-latent compression as the key factor that limits effective scaling. To address this issue, we decompose token-to-latent compression into two stages, reducing structural information loss and enabling effective token number scaling for generation. Secondly, to further mitigate latent representation collapse, we enhance the semantic structure of image tokens via joint self-supervised training, leading to more generative-friendly latents. With these designs, TC-AE achieves substantially improved reconstruction and generative performance under deep compression. We hope our research will advance ViT-based tokenizer for visual generation.

cs.CV↗
arXiv2026

ORION: Unifying Top-Down and Bottom-Up Chemical Space Sampling for a Universal Organic Force Field

Empirical force fields remain the primary tool for large-scale molecular simulation, yet their limited flexibility and transferability often hinder predictive modeling in chemically complex condensed-phase systems. Here we present ORION, a universal machine-learning force field for C, H, O, N, S, and P systems developed within the Neuroevolution Potential (NEP) framework. To enhance transferability across diverse chemical environments, ORION was trained on a chemically rich dataset constructed through an integrated top-down and bottom-up strategy, enabling accurate descriptions of complex organic configurations, reactive intermediates, and weak intermolecular interactions. ORION achieves near-density-functional-theory accuracy while retaining the efficiency required for large-scale molecular dynamics simulations. On the test set, it predicts atomic forces with substantially higher accuracy than ReaxFF while running 215.5 times faster under identical hardware conditions, making simulations on the hundreds-of-nanoseconds timescale readily accessible. The model provides a balanced description of bond breaking and formation, aromatic growth, hydrogen bonding, van der Waals interactions, and π-stacking, demonstrating strong transferability across both reactive and nonreactive systems. These results establish ORION as a practical and general force field for predictive simulations in chemistry and materials science, and provide an effective route toward universal machine-learning force fields with both high accuracy and broad applicability.

physics.chem-ph↗
arXiv2026

Behavioral Generative Agents for Power Dispatch and Auction

This paper presents positive initial evidence that generative agents can relax the rigidity of traditional mathematical models for human decision-making in power dispatch and auction settings. We design two proof-of-concept energy experiments with generative agents powered by a large language model (LLM). First, we construct a home battery management testbed with stochastic electricity prices and blackout interventions, and benchmark LLM decisions against dynamic programming. By incorporating an in-context learning (ICL) module, we show that behavioral patterns discovered by a stronger reasoning model can be transferred to a smaller LLM via example-based prompting, leading agents to prioritize post-blackout energy reserves over short-term profit. Second, we study LLM agents in simultaneous ascending auctions (SAA) for power network access, comparing their behavior with an optimization benchmark, the straightforward bidding strategy. By designing ICL prompts with rule-based, myopic, and strategic objectives, we find that structured prompting combined with ICL enables LLM agents to both reproduce economically rational strategies and exhibit systematic behavioral deviations. Overall, these results suggest that LLM-powered agents provide a flexible and expressive testbed for modeling human decision-making in power system applications.

eess.SY↗
arXiv2026

Behavioral Generative Agents for Energy Operations

Problem definition: Accurately modeling consumer behavior in energy operations is challenging due to uncertainty, behavioral heterogeneity, and limited empirical data-particularly in low-frequency, high-impact events. While generative AI trained on large-scale human data offers new opportunities to study decision behavior, its role in operational applications remains unclear. We examine how generative agents can support customer behavior discovery in energy operations, complementing rather than replacing human-based experiments. Methodology/results: We introduce a novel approach leveraging generative agents-artificial agents powered by large language models-to simulate sequential customer decisions under dynamic electricity prices and outage risks. We find that these agents behave more optimally and rationally in simpler market scenarios, while their performance becomes more variable and suboptimal as task complexity rises. Furthermore, the agents exhibit heterogeneous customer preferences, consistently maintaining distinct, persona-driven reasoning patterns in both operational decisions and textual reasoning. Comparisons with dynamic programming and greedy policy benchmarks show alignment between specific personas and distinct heuristic decision policies. In low-frequency, high-impact events such as blackouts, agents prioritize energy reliability over cost or profit, demonstrating their ability to uncover behavioral patterns beyond the rigidity of traditional mathematical models. Managerial Implications: Our findings suggest that behavioral generative agents can serve as scalable and flexible tools for studying consumer behavior in energy operations. By enabling controlled experiments across heterogeneous customer types and rare events, these agents can enhance the design of energy management systems and support more informed analysis of energy policies and incentive programs.

cs.AI↗
arXiv2026

LongCat-Flash-Thinking-2601 Technical Report

We introduce LongCat-Flash-Thinking-2601, a 560-billion-parameter open-source Mixture-of-Experts (MoE) reasoning model with superior agentic reasoning capability. LongCat-Flash-Thinking-2601 achieves state-of-the-art performance among open-source models on a wide range of agentic benchmarks, including agentic search, agentic tool use, and tool-integrated reasoning. Beyond benchmark performance, the model demonstrates strong generalization to complex tool interactions and robust behavior under noisy real-world environments. Its advanced capability stems from a unified training framework that combines domain-parallel expert training with subsequent fusion, together with an end-to-end co-design of data construction, environments, algorithms, and infrastructure spanning from pre-training to post-training. In particular, the model's strong generalization capability in complex tool-use are driven by our in-depth exploration of environment scaling and principled task construction. To optimize long-tailed, skewed generation and multi-turn agentic interactions, and to enable stable training across over 10,000 environments spanning more than 20 domains, we systematically extend our asynchronous reinforcement learning framework, DORA, for stable and efficient large-scale multi-environment training. Furthermore, recognizing that real-world tasks are inherently noisy, we conduct a systematic analysis and decomposition of real-world noise patterns, and design targeted training procedures to explicitly incorporate such imperfections into the training process, resulting in improved robustness for real-world applications. To further enhance performance on complex reasoning tasks, we introduce a Heavy Thinking mode that enables effective test-time scaling by jointly expanding reasoning depth and width through intensive parallel thinking.

cs.AI↗
arXiv2026

Two-dose vs. Three-Dose Optimization Under Sample Size Constraint

Dose optimization is a hallmark of Project Optimus for oncology drug development. The number of doses to include in a dose optimization study depends on the totality of evidence, which is often unclear in early-phase development. With equal sample sizes per dose, carrying three doses is clearly more advantageous than two for optimization. In this paper, we show that, even when the total sample size is fixed, it is still preferable to carry three unless there is very strong evidence that one can be dropped. A mathematical approximation is applied to guide the investigation, followed by a simulation study to complement the theoretical findings. Semi-quantitative guidance is provided for practitioners, addressing both randomized and non-randomized dose optimization while considering population homogeneity.

stat.ME↗
arXiv2026

AviationLMM: A Large Multimodal Foundation Model for Civil Aviation

Civil aviation is a cornerstone of global transportation and commerce, and ensuring its safety, efficiency and customer satisfaction is paramount. Yet conventional Artificial Intelligence (AI) solutions in aviation remain siloed and narrow, focusing on isolated tasks or single modalities. They struggle to integrate heterogeneous data such as voice communications, radar tracks, sensor streams and textual reports, which limits situational awareness, adaptability, and real-time decision support. This paper introduces the vision of AviationLMM, a Large Multimodal foundation Model for civil aviation, designed to unify the heterogeneous data streams of civil aviation and enable understanding, reasoning, generation and agentic applications. We firstly identify the gaps between existing AI solutions and requirements. Secondly, we describe the model architecture that ingests multimodal inputs such as air-ground voice, surveillance, on-board telemetry, video and structured texts, and performs cross-modal alignment and fusion, and produces flexible outputs ranging from situation summaries and risk alerts to predictive diagnostics and multimodal incident reconstructions. In order to fully realize this vision, we identify key research opportunities to address, including data acquisition, alignment and fusion, pretraining, reasoning, trustworthiness, privacy, robustness to missing modalities, and synthetic scenario generation. By articulating the design and challenges of AviationLMM, we aim to boost the civil aviation foundation model progress and catalyze coordinated research efforts toward an integrated, trustworthy and privacy-preserving aviation AI ecosystem.

cs.AI↗
arXiv2025

Introducing LongCat-Flash-Thinking: A Technical Report

We present LongCat-Flash-Thinking, an efficient 560-billion-parameter open-source Mixture-of-Experts (MoE) reasoning model. Its advanced capabilities are cultivated through a meticulously crafted training process, beginning with long Chain-of-Thought (CoT) data cold-start and culminating in large-scale Reinforcement Learning (RL). We first employ a well-designed cold-start training strategy, which significantly enhances the reasoning potential and equips the model with specialized skills in both formal and agentic reasoning. Then, a core innovation is our domain-parallel training scheme, which decouples optimization across distinct domains (e.g., STEM, Code, Agentic) and subsequently fuses the resulting expert models into a single, nearly Pareto-optimal model. This entire process is powered by our Dynamic ORchestration for Asynchronous rollout (DORA) system, a large-scale RL framework that delivers a greater than threefold training speedup over synchronous methods on tens of thousands of accelerators. As a result, LongCat-Flash-Thinking achieves state-of-the-art performance among open-source models on a suite of complex reasoning tasks. The model exhibits exceptional efficiency in agentic reasoning, reducing average token consumption by 64.5% (from 19, 653 to 6, 965) on AIME-25, without degrading task accuracy. We release LongCat-Flash-Thinking to promote further advances in reasoning systems and agentic AI research.

cs.AI↗
arXiv2025

Dyson Trace Flow and Multivariate Dynamic Coupled Semicircle Law

Interacting random matrix systems are fundamental to modern theoretical physics and data science, yet a unified framework for their analysis has been lacking. This work introduces such a universal framework, built upon two novel concepts: the Dyson Trace Flow characterizing macroscopic fluctuations, and the Multivariate Dynamic Coupled Semicircle Law describing the collective spectral behavior of multiple interacting matrix processes. We establish the stochastic evolution of eigenvalues under asymmetric coupling and prove the mathematical well-posedness of the theory. A large deviation principle is derived, enabling the calculation of rare event probabilities. The framework is extended to nonlinear and non-reciprocal interactions, revealing universal phenomena including exceptional points, bistability, and novel scaling laws. A striking connection to quantum chaos is unveiled through a holographic correspondence with wormhole geometries. By generalizing classical random matrix theory, this work provides powerful tools for understanding neural networks and complex quantum dynamics.

math.PR↗
arXiv2025

Communications to Circulations: Real-Time 3D Wind Field Prediction Using 5G GNSS Signals and Deep Learning

Accurate atmospheric wind field information is crucial for various applications, including weather forecasting, aviation safety, and disaster risk reduction. However, obtaining high spatiotemporal resolution wind data remains challenging due to limitations in traditional in-situ observations and remote sensing techniques, as well as the computational expense and biases of numerical weather prediction (NWP) models. This paper introduces G-WindCast, a novel deep learning framework that leverages signal strength variations from 5G Global Navigation Satellite System (GNSS) signals to forecast three-dimensional (3D) atmospheric wind fields. The framework utilizes Forward Neural Networks (FNN) and Transformer networks to capture complex, nonlinear, and spatiotemporal relationships between GNSS-derived features and wind dynamics. Our preliminary results demonstrate promising accuracy in real-time wind forecasts (up to 30 minutes lead time). The model exhibits robustness across forecast horizons and different pressure levels, and its predictions for wind fields show superior agreement with ground-based radar wind profiler compared to concurrent European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis v5 (ERA5). Furthermore, we show that the system can maintain excellent performance for localized forecasting even with a significantly reduced number of GNSS stations (e.g., around 100), highlighting its cost-effectiveness and scalability. This interdisciplinary approach underscores the transformative potential of exploiting non-traditional data sources and deep learning for advanced environmental monitoring and real-time atmospheric applications.

cs.LG↗