arXiv subjects

Ruiyi Wang

Publications and source records attributed to Ruiyi Wang.

17 recordsLinked to original sources

arXiv2026

Parity Mapping for Quantum Optimization on Frustrated Ising Rings

The frustrated Ising ring is one of the simplest models exhibiting exponential closing spectral gaps, making it a paradigmatic and challenging benchmark for quantum annealing (QA). Ground-state preparation for this model has therefore been studied extensively in both continuous-time QA and digitized protocols such as the Quantum Approximate Optimization Algorithm (QAOA). Here, we use the frustrated Ising ring to investigate how the parity mapping affects the performance of both QA and QAOA. For QA, finite-size calculations show that the parity mapping increases the minimum spectral gap under the energy normalization used in this work, thereby enabling faster continuous-time ground state preparation protocols. An ideal implementation of Parity-QA, with a single global constraint, shows no evidence of exponential gap closing over the accessible system sizes, whereas a hardware-motivated decomposition into local constraints restores the exponential decrease, albeit with a smaller fitted exponent than conventional QA. For the digitized protocol, we find that the number of Parity-QAOA layers required to prepare the exact ground state remains constant over the simulated sizes, improving upon the quadratic scaling required by conventional QAOA. To investigate the role of constraints in Parity-QAOA, we further consider a modified Ising ring instance in which the constraint term is essential for preparing the target ground state. We then compare the corresponding resource requirements with those of conventional QAOA.

quant-ph↗

arXiv2026

How to Instruct Your Robot: Dense Language Annotations Power Robot Policy Learning

Scaling robot policy learning is bottlenecked by the cost of collecting demonstrations, while language annotations for existing demonstrations are comparatively cheap. We study language density as a lever for extracting more signal from a fixed robot or egocentric-video corpus. We introduce DeMiAn (Dense Multi-aspect Annotation), a two-stage approach that first re-labels demonstration segments with VLM-generated annotations along four complementary aspects: physical motion, scene composition, arm pose, and reasoning. A learned instructor then maps a task description and initial scene snapshot to a task-appropriate annotation at deployment, running asynchronously so generation latency is hidden behind policy execution. Across over 1M robot manipulation clips and 50K EgoVerse human-egocentric videos, DeMiAn improves both a vision-language-action policy and a video-based world-action model without collecting new demonstrations. On RoboCasa, the instructor raises success by 5 points over a task-only baseline and comes within 3 points of a per-task oracle. No fixed annotation aspect dominates across tasks, showing that selecting the right dense language matters. DeMiAn also improves composite-task and out-of-distribution performance, and shifts the compute-performance frontier in both mid-training and post-training after accounting for annotation-generation FLOPs. These results position dense re-annotation as a practical scaling lever for robot policy learning.

cs.RO↗

arXiv2026

ChipMATE: Multi-Agent Training via Reinforcement Learning for Enhanced RTL Generation

Existing API-based agentic systems for RTL code generation are fundamentally misaligned with industrial practice: they assume a golden testbench is available at generation time, rely on closed-source APIs incompatible with chip vendors' air-gapped security requirements, and cannot be trained on vendors' proprietary RTL codebases, leaving valuable internal data unused. Recent self-trained models address the deployment constraint but remain single-turn generators that overlook the critical role of verification in real industrial flows. To bridge these gaps, we present ChipMATE, the first self-trained multi-agent framework for RTL generation. Inspired by industrial practice where correctness emerges from cross-comparison between independently written RTL modules and reference models, ChipMATE pairs a Verilog agent with a Python reference-model agent that mutually verify each other's outputs without any golden oracle. We design a backtrack-based inference workflow to prevent error propagation across turns, and a two-stage training pipeline that first trains each agent individually to saturate its code-generation capability, then trains the team jointly to collaborate effectively. To support the training, we further build a hybrid data-generation framework that produces 64.4K high-quality reference model training samples. ChipMATE achieves 75.0\% and 80.1\% pass@1 on VerilogEval V2 with 4B and 9B base models, outperforming all existing self-trained models and even DeepSeek V4 with 1600B parameters. Our code and model weights are publicly available in https://github.com/zhongkaiyu/ChipMATE.

cs.MA↗

arXiv2026

FlashEvolve: Accelerating Agent Self-Evolution with Asynchronous Stage Orchestration

LLM-based evolution has emerged as a promising way to improve agents by refining non-parametric artifacts, but its wall-clock cost remains a major bottleneck. We identify that this cost comes from synchronized stage execution and imbalance inside each LLM-heavy stage. We present FlashEvolve, an efficient framework that replaces synchronized execution with asynchronous workers and queues, allowing different stages and steps to overlap. To handle data staleness introduced by asynchrony, FlashEvolve tracks artifact versions and applies different policies to update, discard, or patch stale artifacts. Unlike weight-space staleness in asynchronous RL, language-space staleness is inspectable and repairable: a stale artifact is not just delayed work, but readable evidence that the LLM can reflect on, revise, and turn into useful evolution signal. FlashEvolve further improves throughput and token efficiency with speculative stage completion and adaptive workflow control. On GEPA workloads, FlashEvolve improves proposal throughput by $3.5\times$ on local vLLM and $4.9\times$ on API serving over synchronous GEPA. The same design also applies to ACE and Meta-Harness.

cs.LG↗

arXiv2025

A Practitioner's Guide to Multi-turn Agentic Reinforcement Learning

We study what actually works and what doesn't for training large language models as agents via multi-turn reinforcement learning. Despite rapid progress, existing frameworks and definitions are fragmented, and there is no systematic formulation or analysis of which design choices matter across tasks. We address this gap by first breaking down the design space into three inter-related pillars -- environment, reward, and policy -- and empirically derive a recipe for training LLM agents in situated textual domains. In particular, we test TextWorld and ALFWorld, popular domains for testing situated embodied reasoning, as well as SWE-Gym for more software engineering style tasks. (i) For the environment, we analyze the impacts of task complexity in terms of sizes of the state and action spaces as well as optimal solution length, finding that even simple environments within a domain can provide signal on how well an agent can generalize to more complex tasks. (ii) For the reward, we ablate relative reward sparsity, observing that while dense turn-level rewards accelerate training, performance and stability is highly dependent on the choice of RL algorithm. (iii) And for the agent's policy, we explore the interplay between reward sparsity and biased (PPO, GRPO) and unbiased (RLOO) policy gradient methods in addition to showing how to find the optimal Supervised Fine-tuning (SFT) to RL training ratio given a fixed budget. We distill these findings into a training recipe that guides co-design across the three pillars, facilitating research and practical efforts in multi-turn agentic RL. Code: https://github.com/pearls-lab/meow-tea-taro

cs.LG↗

arXiv2025

Digital controllability of transverse field Ising chains

Quantum Annealing (QA) encounters limitations when the energy gap along the annealing path becomes exponentially small, leading to impractically long runtimes. In contrast, the success of hybrid digital methods like the Quantum Approximate Optimization Algorithm (QAOA), which operate via discrete unitary operations, relies on the optimization of the variational parameters appearing in the state. We analyze a class of transverse-field Ising models which includes problems with exponentially small spectral gaps, but whose dynamics is described in terms of fermionic Gaussian states after Jordan-Wigner mapping. We show that, for digital alternating QAOA-like states, the number of unitaries required to reach the exact ground state scales quadratically with system size and is independent of the annealing gap. This number can be exactly computed from the algebraic properties of the Ansatz, revealing a fundamental distinction between digital methods and their analog counterpart.

quant-ph↗

arXiv2025

DreamPRM: Domain-Reweighted Process Reward Model for Multimodal Reasoning

Reasoning has substantially improved the performance of large language models (LLMs) on complicated tasks. Central to the current reasoning studies, Process Reward Models (PRMs) offer a fine-grained evaluation of intermediate reasoning steps and guide the reasoning process. However, extending PRMs to multimodal large language models (MLLMs) introduces challenges. Since multimodal reasoning covers a wider range of tasks compared to text-only scenarios, the resulting distribution shift from the training to testing sets is more severe, leading to greater generalization difficulty. Training a reliable multimodal PRM, therefore, demands large and diverse datasets to ensure sufficient coverage. However, current multimodal reasoning datasets suffer from a marked quality imbalance, which degrades PRM performance and highlights the need for an effective data selection strategy. To address the issues, we introduce DreamPRM, a domain-reweighted training framework for multimodal PRMs which employs bi-level optimization. In the lower-level optimization, DreamPRM performs fine-tuning on multiple datasets with domain weights, allowing the PRM to prioritize high-quality reasoning signals and alleviating the impact of dataset quality imbalance. In the upper-level optimization, the PRM is evaluated on a separate meta-learning dataset; this feedback updates the domain weights through an aggregation loss function, thereby improving the generalization capability of trained PRM. Extensive experiments on multiple multimodal reasoning benchmarks covering both mathematical and general reasoning show that test-time scaling with DreamPRM consistently improves the performance of state-of-the-art MLLMs. Further comparisons reveal that DreamPRM's domain-reweighting strategy surpasses other data selection methods and yields higher accuracy gains than existing test-time scaling approaches.

cs.LG↗

arXiv2025

Learning Hazing to Dehazing: Towards Realistic Haze Generation for Real-World Image Dehazing

Existing real-world image dehazing methods primarily attempt to fine-tune pre-trained models or adapt their inference procedures, thus heavily relying on the pre-trained models and associated training data. Moreover, restoring heavily distorted information under dense haze requires generative diffusion models, whose potential in dehazing remains underutilized partly due to their lengthy sampling processes. To address these limitations, we introduce a novel hazing-dehazing pipeline consisting of a Realistic Hazy Image Generation framework (HazeGen) and a Diffusion-based Dehazing framework (DiffDehaze). Specifically, HazeGen harnesses robust generative diffusion priors of real-world hazy images embedded in a pre-trained text-to-image diffusion model. By employing specialized hybrid training and blended sampling strategies, HazeGen produces realistic and diverse hazy images as high-quality training data for DiffDehaze. To alleviate the inefficiency and fidelity concerns associated with diffusion-based methods, DiffDehaze adopts an Accelerated Fidelity-Preserving Sampling process (AccSamp). The core of AccSamp is the Tiled Statistical Alignment Operation (AlignOp), which can provide a clean and faithful dehazing estimate within a small fraction of sampling steps to reduce complexity and enable effective fidelity guidance. Extensive experiments demonstrate the superior dehazing performance and visual quality of our approach over existing methods. The code is available at https://github.com/ruiyi-w/Learning-Hazing-to-Dehazing.

cs.CV↗

arXiv2025

From Exponential to Quadratic: Optimal Control for a Frustrated Ising Ring Model

Exponentially small spectral gaps are known to be the crucial bottleneck for traditional Quantum Annealing (QA) based on interpolating between two Hamiltonians, a simple driving term and the complex problem to be solved, with a linear schedule in time. One of the simplest models showing exponentially small spectral gaps was introduced by Roberts et al., PRA 101, 042317 (2020): a ferromagnetic Ising ring with a single frustrating antiferromagnetic bond. A previous study of this model (C\^ot\'e et al., QST 8, 045033 (2023)) proposed a continuous-time diabatic QA, where optimized non-adiabatic annealing schedules provided good solutions, avoiding exponentially large annealing times. In our work, we move to a digital framework of Variational Quantum Algorithms, and present two main results: 1) we show that the model is digitally controllable with a scaling of resources that grows quadratically with the system size, achieving the exact solution using the Quantum Approximate Optimization Algorithm (QAOA); 2) We combine a technique of quantum control -- the Chopped RAndom Basis (CRAB) method -- and digitized quantum annealing (dQA) to construct smooth digital schedules yielding optimal solutions with very high accuracy.

quant-ph↗

arXiv2024

HazeCLIP: Towards Language Guided Real-World Image Dehazing

Existing methods have achieved remarkable performance in image dehazing, particularly on synthetic datasets. However, they often struggle with real-world hazy images due to domain shift, limiting their practical applicability. This paper introduces HazeCLIP, a language-guided adaptation framework designed to enhance the real-world performance of pre-trained dehazing networks. Inspired by the Contrastive Language-Image Pre-training (CLIP) model's ability to distinguish between hazy and clean images, we leverage it to evaluate dehazing results. Combined with a region-specific dehazing technique and tailored prompt sets, the CLIP model accurately identifies hazy areas, providing a high-quality, human-like prior that guides the fine-tuning process of pre-trained networks. Extensive experiments demonstrate that HazeCLIP achieves state-of-the-art performance in real-word image dehazing, evaluated through both visual quality and image quality assessment metrics. Codes are available at https://github.com/Troivyn/HazeCLIP.

cs.CV↗

arXiv2024

PATIENT-{\Psi}: Using Large Language Models to Simulate Patients for Training Mental Health Professionals

Mental illness remains one of the most critical public health issues. Despite its importance, many mental health professionals highlight a disconnect between their training and actual real-world patient practice. To help bridge this gap, we propose PATIENT-{\Psi}, a novel patient simulation framework for cognitive behavior therapy (CBT) training. To build PATIENT-{\Psi}, we construct diverse patient cognitive models based on CBT principles and use large language models (LLMs) programmed with these cognitive models to act as a simulated therapy patient. We propose an interactive training scheme, PATIENT-{\Psi}-TRAINER, for mental health trainees to practice a key skill in CBT -- formulating the cognitive model of the patient -- through role-playing a therapy session with PATIENT-{\Psi}. To evaluate PATIENT-{\Psi}, we conducted a comprehensive user study of 13 mental health trainees and 20 experts. The results demonstrate that practice using PATIENT-{\Psi}-TRAINER enhances the perceived skill acquisition and confidence of the trainees beyond existing forms of training such as textbooks, videos, and role-play with non-patients. Based on the experts' perceptions, PATIENT-{\Psi} is perceived to be closer to real patient interactions than GPT-4, and PATIENT-{\Psi}-TRAINER holds strong promise to improve trainee competencies. Our code and data are released at \url{https://github.com/ruiyiw/patient-psi}.

cs.CL↗

arXiv2024

DehazeDCT: Towards Effective Non-Homogeneous Dehazing via Deformable Convolutional Transformer

Image dehazing, a pivotal task in low-level vision, aims to restore the visibility and detail from hazy images. Many deep learning methods with powerful representation learning capability demonstrate advanced performance on non-homogeneous dehazing, however, these methods usually struggle with processing high-resolution images (e.g., $4000 \times 6000$) due to their heavy computational demands. To address these challenges, we introduce an innovative non-homogeneous Dehazing method via Deformable Convolutional Transformer-like architecture (DehazeDCT). Specifically, we first design a transformer-like network based on deformable convolution v4, which offers long-range dependency and adaptive spatial aggregation capabilities and demonstrates faster convergence and forward speed. Furthermore, we leverage a lightweight Retinex-inspired transformer to achieve color correction and structure refinement. Extensive experiment results and highly competitive performance of our method in NTIRE 2024 Dense and Non-Homogeneous Dehazing Challenge, ranking second among all 16 submissions, demonstrate the superior capability of our proposed method. The code is available: https://github.com/movingforward100/Dehazing_R.

cs.CV↗

arXiv2024

Beyond Quantum Annealing: Optimal control solutions to MaxCut problems

Quantum Annealing (QA) relies on mixing two Hamiltonian terms, a simple driver and a complex problem Hamiltonian, in a linear combination. The time-dependent schedule for this mixing is often taken to be linear in time: improving on this linear choice is known to be essential and has proven to be difficult. Here, we present different techniques for improving on the linear-schedule QA along two directions, conceptually distinct but leading to similar outcomes: 1) the first approach consists of constructing a Trotter-digitized QA (dQA) with schedules parameterized in terms of Fourier modes or Chebyshev polynomials, inspired by the Chopped Random Basis algorithm (CRAB) for optimal control in continuous time; 2) the second approach is technically a Quantum Approximate Optimization Algorithm (QAOA), whose solutions are found iteratively using linear interpolation or expansion in Fourier modes. Both approaches emphasize finding smooth optimal schedule parameters, ultimately leading to hybrid quantum-classical variational algorithms of the alternating Hamiltonian Ansatz type. We apply these techniques to MaxCut problems on weighted 3-regular graphs with N = 14 sites, focusing on hard instances that exhibit a small spectral gap, for which a standard linear-schedule QA performs poorly. We characterize the physics behind the optimal protocols for both the dQA and QAOA approaches, discovering shortcuts to adiabaticity-like dynamics. Furthermore, we study the transferability of such smooth solutions among hard instances of MaxCut at different circuit depths. Finally, we show that the smoothness pattern of these protocols obtained in a digital setting enables us to adapt them to continuous-time evolution, contrarily to generic non-smooth solutions. This procedure results in an optimized quantum annealing schedule that is implementable on analog devices.

quant-ph↗

arXiv2024

SOTOPIA-$π$: Interactive Learning of Socially Intelligent Language Agents

Humans learn social skills through both imitation and social interaction. This social learning process is largely understudied by existing research on building language agents. Motivated by this gap, we propose an interactive learning method, SOTOPIA-$π$, improving the social intelligence of language agents. This method leverages behavior cloning and self-reinforcement training on filtered social interaction data according to large language model (LLM) ratings. We show that our training method allows a 7B LLM to reach the social goal completion ability of an expert model (GPT-4-based agent), while improving the safety of language agents and maintaining general QA ability on the MMLU benchmark. We also find that this training paradigm uncovers some difficulties in LLM-based evaluation of social intelligence: LLM-based evaluators overestimate the abilities of the language agents trained specifically for social interaction.

cs.CL↗

arXiv2024

Missing Pieces: How Do Designs that Expose Uncertainty Longitudinally Impact Trust in AI Decision Aids? An In Situ Study of Gig Drivers

Decision aids based on artificial intelligence (AI) induce a wide range of outcomes when they are deployed in uncertain environments. In this paper, we investigate how users' trust in recommendations from an AI decision aid is impacted over time by designs that expose uncertainty in predicted outcomes. Unlike previous work, we focus on gig driving - a real-world, repeated decision-making context. We report on a longitudinal mixed-methods study ($n=51$) where we measured gig drivers' trust as they interacted with an AI-based schedule recommendation tool. Our results show that participants' trust in the tool was shaped by both their first impressions of its accuracy and their longitudinal interactions with it; and that task-aligned framings of uncertainty improved trust by allowing participants to incorporate uncertainty into their decision-making processes. Additionally, we observed that trust depended on their characteristics as drivers, underscoring the need for more in situ studies of AI decision aids.

cs.HC↗

arXiv2024

"I'm categorizing LLM as a productivity tool": Examining ethics of LLM use in HCI research practices

Large language models are increasingly applied in real-world scenarios, including research and education. These models, however, come with well-known ethical issues, which may manifest in unexpected ways in human-computer interaction research due to the extensive engagement with human subjects. This paper reports on research practices related to LLM use, drawing on 16 semi-structured interviews and a survey conducted with 50 HCI researchers. We discuss the ways in which LLMs are already being utilized throughout the entire HCI research pipeline, from ideation to system development and paper writing. While researchers described nuanced understandings of ethical issues, they were rarely or only partially able to identify and address those ethical concerns in their own projects. This lack of action and reliance on workarounds was explained through the perceived lack of control and distributed responsibility in the LLM supply chain, the conditional nature of engaging with ethics, and competing priorities. Finally, we reflect on the implications of our findings and present opportunities to shape emerging norms of engaging with large language models in HCI research.

cs.HC↗

arXiv2023

Long-Horizon Dialogue Understanding for Role Identification in the Game of Avalon with Large Language Models

Deception and persuasion play a critical role in long-horizon dialogues between multiple parties, especially when the interests, goals, and motivations of the participants are not aligned. Such complex tasks pose challenges for current Large Language Models (LLM) as deception and persuasion can easily mislead them, especially in long-horizon multi-party dialogues. To this end, we explore the game of Avalon: The Resistance, a social deduction game in which players must determine each other's hidden identities to complete their team's objective. We introduce an online testbed and a dataset containing 20 carefully collected and labeled games among human players that exhibit long-horizon deception in a cooperative-competitive setting. We discuss the capabilities of LLMs to utilize deceptive long-horizon conversations between six human players to determine each player's goal and motivation. Particularly, we discuss the multimodal integration of the chat between the players and the game's state that grounds the conversation, providing further insights into the true player identities. We find that even current state-of-the-art LLMs do not reach human performance, making our dataset a compelling benchmark to investigate the decision-making and language-processing capabilities of LLMs. Our dataset and online testbed can be found at our project website: https://sstepput.github.io/Avalon-NLU/

cs.CL↗