arXiv subjects

Tian Xia

Publications and source records attributed to Tian Xia.

At least 73 records · Page 4Linked to original sources

arXiv2025

Response-Level Rewards Are All You Need for Online Reinforcement Learning in LLMs: A Mathematical Perspective

We study a common challenge in reinforcement learning for large language models (LLMs): the Zero-Reward Assumption, where non-terminal actions (i.e., intermediate token generations) receive zero task-specific immediate reward, while only the final token receives a reward for the entire response. This assumption arises frequently in practice, as precise token-level rewards are often difficult or infeasible to obtain in LLM applications. In this work, we provide a unifying theoretical perspective. We introduce the Trajectory Policy Gradient Theorem, which shows that the policy gradient based on true, unknown token-level rewards can be unbiasedly estimated using only a response-level reward model, regardless of whether the Zero-Reward Assumption holds or not, for algorithms in the REINFORCE and Actor-Critic families. This result reveals that widely used methods such as PPO, GRPO, ReMax, and RLOO inherently possess the capacity to model token-level reward signals, offering a theoretical justification for response-level reward approaches. Our findings pave the way for more practical, efficient LLM fine-tuning, allowing developers to treat training algorithms as black boxes and focus on improving the response-level reward model with auxiliary sub-models. We also offer a detailed analysis of popular RL and non-RL methods, comparing their theoretical foundations and practical advantages across common LLM tasks. Finally, we propose a new algorithm: Token-Reinforced Policy Optimization (TRePO), a theoretically grounded method that is simpler than PPO, matches GRPO in memory efficiency, and holds promise for broad applicability.

cs.LG↗

arXiv2025

High coherence fluxonium manufactured with a wafer-scale uniformity process

Fluxonium qubits are recognized for their high coherence times and high operation fidelities, attributed to their unique design incorporating a superinductor, which is typically implemented using an array of over 100 Josephson junctions. However, this complexity poses significant fabrication challenges, particularly in achieving high yield and junction uniformity with traditional methods. Here, we introduce an overlap process for Josephson junction fabrication that achieves nearly 100% yield and maintains uniformity across a 2-inch wafer with less than 5% variation for the phase slip junction and less than 2% for the entire junction array. We use a compact junction array design that achieves state-of-the-art dielectric loss tangents and flux noise levels, as confirmed by multiple devices. This enables fluxonium qubits to reach energy relaxation times exceeding 1 millisecond at the flux frustration point. This work paves the way for scalable high coherence fluxonium processors using CMOS-compatible processes, marking a significant step towards practical quantum computing.

quant-ph↗

arXiv2025

High-Dimensional Light Field Modulation via Conjugate Phase Encoding in Liquid Crystal Devices

High-dimensional light field modulation demands precise control over multiple optical parameters, a capability critical for next-generation photonic systems. While liquid crystals offer inherent advantages in dynamic birefringence tuning, existing approaches face fundamental limitations in decoupling interdependent phase responses across polarization states. Here, we demonstrate a conjugate phase encoding paradigm enabling simultaneous manipulation of wavelength-dependent wavefronts, orbital angular momentum (OAM), and polarization states in photoaligned liquid crystal devices.

physics.optics↗

arXiv2025

The effects of disorder in superconducting materials on qubit coherence

Introducing disorderness in the superconducting materials has been considered promising to enhance the electromagnetic impedance and realize noise-resilient superconducting qubits. Despite a number of pioneering implementations, the understanding of the correlation between the material disorderness and the qubit coherence is still developing. Here, we demonstrate a systematic characterization of fluxonium qubits with the superinductors made from titanium-aluminum-nitride with varied disorderness. From qubit noise spectroscopy, the flux noise and the dielectric loss are extracted as a measure of the coherence properties. Our results reveal that the $1/f$ flux noise dominates the qubit decoherence around the flux-frustration point, strongly correlated with the material disorderness; while the dielectric loss remains low under a wide range of material properties. From the flux-noise amplitudes, the areal density ($σ$) of the phenomenological spin defects and material disorderness are found to be approximately correlated by $σ\propto ρ_{xx}^3$, or effectively $(k_F l)^{-3}$. This work has provided new insights on the origin of decoherence channels within superconductors, and could serve as a useful guideline for material design and optimization.

quant-ph↗

arXiv2025

Robust image representations with counterfactual contrastive learning

Contrastive pretraining can substantially increase model generalisation and downstream performance. However, the quality of the learned representations is highly dependent on the data augmentation strategy applied to generate positive pairs. Positive contrastive pairs should preserve semantic meaning while discarding unwanted variations related to the data acquisition domain. Traditional contrastive pipelines attempt to simulate domain shifts through pre-defined generic image transformations. However, these do not always mimic realistic and relevant domain variations for medical imaging, such as scanner differences. To tackle this issue, we herein introduce counterfactual contrastive learning, a novel framework leveraging recent advances in causal image synthesis to create contrastive positive pairs that faithfully capture relevant domain variations. Our method, evaluated across five datasets encompassing both chest radiography and mammography data, for two established contrastive objectives (SimCLR and DINO-v2), outperforms standard contrastive learning in terms of robustness to acquisition shift. Notably, counterfactual contrastive learning achieves superior downstream performance on both in-distribution and external datasets, especially for images acquired with scanners under-represented in the training set. Further experiments show that the proposed framework extends beyond acquisition shifts, with models trained with counterfactual contrastive learning reducing subgroup disparities across biological sex.

cs.CV↗

arXiv2025

Systematic Evaluation of LLM-as-a-Judge in LLM Alignment Tasks: Explainable Metrics and Diverse Prompt Templates

LLM-as-a-Judge has been widely applied to evaluate and compare different LLM alignmnet approaches (e.g., RLHF and DPO). However, concerns regarding its reliability have emerged, due to LLM judges' biases and inconsistent decision-making. Previous research has developed evaluation frameworks to assess reliability of LLM judges and their alignment with human preferences. However, the employed evaluation metrics often lack adequate explainability and fail to address LLM internal inconsistency. Additionally, existing studies inadequately explore the impact of various prompt templates when applying LLM-as-a-Judge methods, leading to potentially inconsistent comparisons between different alignment algorithms. In this work, we systematically evaluate LLM-as-a-Judge on alignment tasks by defining more theoretically interpretable evaluation metrics and explicitly mitigating LLM internal inconsistency from reliability metrics. We develop an open-source framework to evaluate, compare, and visualize the reliability and alignment of LLM judges, which facilitates practitioners to choose LLM judges for alignment tasks. In the experiments, we examine effects of diverse prompt templates on LLM-judge reliability and also demonstrate our developed framework by comparing various LLM judges on two common alignment datasets (i.e., TL;DR Summarization and HH-RLHF-Helpfulness). Our results indicate a significant impact of prompt templates on LLM judge performance, as well as a mediocre alignment level between the tested LLM judges and human evaluators.

cs.CL↗

arXiv2025

SkyServe: Serving AI Models across Regions and Clouds with Spot Instances

Recent years have witnessed an explosive growth of AI models. The high cost of hosting AI services on GPUs and their demanding service requirements, make it timely and challenging to lower service costs and guarantee service quality. While spot instances have long been offered with a large discount, spot preemptions have discouraged users from using them to host model replicas when serving AI models. To address this, we propose a simple yet efficient policy, SpotHedge, that leverages spot replicas across different failure domains (e.g., regions and clouds) to ensure availability, lower costs, and high service quality. SpotHedge intelligently spreads spot replicas across different regions and clouds to improve availability and reduce correlated preemptions, overprovisions cheap spot replicas than required as a safeguard against possible preemptions, and dynamically falls back to on-demand replicas when spot replicas become unavailable. We built SkyServe, a system leveraging SpotHedge to efficiently serve AI models over a mixture of spot and on-demand replicas across regions and clouds. We compared SkyServe with both research and production systems on real AI workloads: SkyServe reduces cost by 43% on average while achieving high resource availability compared to using on-demand replicas. Additionally, SkyServe improves P50, P90, and P99 latency by 2.3$\times$, 2.1$\times$, 2.1$\times$ on average compared to other research and production systems.

cs.DC↗

arXiv2025

The Danger of Overthinking: Examining the Reasoning-Action Dilemma in Agentic Tasks

Large Reasoning Models (LRMs) represent a breakthrough in AI problem-solving capabilities, but their effectiveness in interactive environments can be limited. This paper introduces and analyzes overthinking in LRMs. A phenomenon where models favor extended internal reasoning chains over environmental interaction. Through experiments on software engineering tasks using SWE Bench Verified, we observe three recurring patterns: Analysis Paralysis, Rogue Actions, and Premature Disengagement. We propose a framework to study these behaviors, which correlates with human expert assessments, and analyze 4018 trajectories. We observe that higher overthinking scores correlate with decreased performance, with reasoning models exhibiting stronger tendencies toward overthinking compared to non-reasoning models. Our analysis reveals that simple efforts to mitigate overthinking in agentic environments, such as selecting the solution with the lower overthinking score, can improve model performance by almost 30% while reducing computational costs by 43%. These results suggest that mitigating overthinking has strong practical implications. We suggest that by leveraging native function-calling capabilities and selective reinforcement learning overthinking tendencies could be mitigated. We also open-source our evaluation framework and dataset to facilitate research in this direction at https://github.com/AlexCuadron/Overthinking.

cs.AI↗

arXiv2025

Nearly Lossless Adaptive Bit Switching

Model quantization is widely applied for compressing and accelerating deep neural networks (DNNs). However, conventional Quantization-Aware Training (QAT) focuses on training DNNs with uniform bit-width. The bit-width settings vary across different hardware and transmission demands, which induces considerable training and storage costs. Hence, the scheme of one-shot joint training multiple precisions is proposed to address this issue. Previous works either store a larger FP32 model to switch between different precision models for higher accuracy or store a smaller INT8 model but compromise accuracy due to using shared quantization parameters. In this paper, we introduce the Double Rounding quantization method, which fully utilizes the quantized representation range to accomplish nearly lossless bit-switching while reducing storage by using the highest integer precision instead of full precision. Furthermore, we observe a competitive interference among different precisions during one-shot joint training, primarily due to inconsistent gradients of quantization scales during backward propagation. To tackle this problem, we propose an Adaptive Learning Rate Scaling (ALRS) technique that dynamically adapts learning rates for various precisions to optimize the training process. Additionally, we extend our Double Rounding to one-shot mixed precision training and develop a Hessian-Aware Stochastic Bit-switching (HASB) strategy. Experimental results on the ImageNet-1K classification demonstrate that our methods have enough advantages to state-of-the-art one-shot joint QAT in both multi-precision and mixed-precision. We also validate the feasibility of our method on detection and segmentation tasks, as well as on LLMs task. Our codes are available at https://github.com/haiduo/Double-Rounding.

cs.CV↗

arXiv2025

Partial Channel Network: Compute Fewer, Perform Better

Designing a module or mechanism that enables a network to maintain low parameters and FLOPs without sacrificing accuracy and throughput remains a challenge. To address this challenge and exploit the redundancy within feature map channels, we propose a new solution: partial channel mechanism (PCM). Specifically, through the split operation, the feature map channels are divided into different parts, with each part corresponding to different operations, such as convolution, attention, pooling, and identity mapping. Based on this assumption, we introduce a novel partial attention convolution (PATConv) that can efficiently combine convolution with visual attention. Our exploration indicates that the PATConv can completely replace both the regular convolution and the regular visual attention while reducing model parameters and FLOPs. Moreover, PATConv can derive three new types of blocks: Partial Channel-Attention block (PAT_ch), Partial Spatial-Attention block (PAT_sp), and Partial Self-Attention block (PAT_sf). In addition, we propose a novel dynamic partial convolution (DPConv) that can adaptively learn the proportion of split channels in different layers to achieve better trade-offs. Building on PATConv and DPConv, we propose a new hybrid network family, named PartialNet, which achieves superior top-1 accuracy and inference speed compared to some SOTA models on ImageNet-1K classification and excels in both detection and segmentation on the COCO dataset. Our code is available at https://github.com/haiduo/PartialNet.

cs.CV↗

arXiv2025

A Modular Program-Transformation Framework for Reducing Specifications to Reachability

Software verification is a complex problem, and verification tools need significant tuning to achieve high performance. Due to this, many verifiers choose to specialize on reachability properties, or invest the time to implement known transformations from the given specification to reachability on their internal representations. To improve this situation, we provide transformations as stand-alone components, modifying the input program instead of the internal representation, enabling their usage as a preprocessing step by other verifiers. This way, we separate two concerns: improving the performance of reachability analyses and implementing efficient transformations of arbitrary specifications to reachability. We implement the transformations in a framework that is based on instrumentation automata, inspired by the BLAST query language. In our initial study, we support three important concrete specifications for C programs: termination, no-overflow, and memory cleanup. Moreover, we discuss the broader expressiveness of our framework and show how general liveness properties can be transformed to reachability. We demonstrate the effectiveness and efficiency of our transformations by comparing verifiers that support the specifications natively with verifiers for reachability applied on the transformed programs. The results are very promising: Our transformations can extend existing verifiers to be effective on specifications that they do not support natively, and that the efficiency is often similar to verifiers that natively support the considered specifications.

cs.PL↗

arXiv2025

Locality-aware Fair Scheduling in LLM Serving

Large language model (LLM) inference workload dominates a wide variety of modern AI applications, ranging from multi-turn conversation to document analysis. Balancing fairness and efficiency is critical for managing diverse client workloads with varying prefix patterns. Unfortunately, existing fair scheduling algorithms for LLM serving, such as Virtual Token Counter (VTC), fail to take prefix locality into consideration and thus suffer from poor performance. On the other hand, locality-aware scheduling algorithms in existing LLM serving frameworks tend to maximize the prefix cache hit rate without considering fair sharing among clients. This paper introduces the first locality-aware fair scheduling algorithm, Deficit Longest Prefix Match (DLPM), which can maintain a high degree of prefix locality with a fairness guarantee. We also introduce a novel algorithm, Double Deficit LPM (D$^2$LPM), extending DLPM for the distributed setup that can find a balance point among fairness, locality, and load-balancing. Our extensive evaluation demonstrates the superior performance of DLPM and D$^2$LPM in ensuring fairness while maintaining high throughput (up to 2.87$\times$ higher than VTC) and low per-client (up to 7.18$\times$ lower than state-of-the-art distributed LLM serving system) latency.

cs.DC↗

arXiv2025

Make Full Use of Testing Information: An Integrated Accelerated Testing and Evaluation Method for Autonomous Driving Systems

Testing and evaluation is an important step before the large-scale application of the autonomous driving systems (ADSs). Based on the three level of scenario abstraction theory, a testing can be performed within a logical scenario, followed by an evaluation stage which is inputted with the testing results of each concrete scenario generated from the logical parameter space. During the above process, abundant testing information is produced which is beneficial for comprehensive and accurate evaluations. To make full use of testing information, this paper proposes an Integrated accelerated Testing and Evaluation Method (ITEM). Based on a Monte Carlo Tree Search (MCTS) paradigm and a dual surrogates testing framework proposed in our previous work, this paper applies the intermediate information (i.e., the tree structure, including the affiliation of each historical sampled point with the subspaces and the parent-child relationship between subspaces) generated during the testing stage into the evaluation stage to achieve accurate hazardous domain identification. Moreover, to better serve this purpose, the UCB calculation method is improved to allow the search algorithm to focus more on the hazardous domain boundaries. Further, a stopping condition is constructed based on the convergence of the search algorithm. Ablation and comparative experiments are then conducted to verify the effectiveness of the improvements and the superiority of the proposed method. The experimental results show that ITEM could well identify the hazardous domains in both low- and high-dimensional cases, regardless of the shape of the hazardous domains, indicating its generality and potential for the safety evaluation of ADSs.

cs.AI↗

arXiv2024

UniCtrl: Improving the Spatiotemporal Consistency of Text-to-Video Diffusion Models via Training-Free Unified Attention Control

Video Diffusion Models have been developed for video generation, usually integrating text and image conditioning to enhance control over the generated content. Despite the progress, ensuring consistency across frames remains a challenge, particularly when using text prompts as control conditions. To address this problem, we introduce UniCtrl, a novel, plug-and-play method that is universally applicable to improve the spatiotemporal consistency and motion diversity of videos generated by text-to-video models without additional training. UniCtrl ensures semantic consistency across different frames through cross-frame self-attention control, and meanwhile, enhances the motion quality and spatiotemporal consistency through motion injection and spatiotemporal synchronization. Our experimental results demonstrate UniCtrl's efficacy in enhancing various text-to-video models, confirming its effectiveness and universality.

cs.CV↗

arXiv2024

R-Judge: Benchmarking Safety Risk Awareness for LLM Agents

Large language models (LLMs) have exhibited great potential in autonomously completing tasks across real-world applications. Despite this, these LLM agents introduce unexpected safety risks when operating in interactive environments. Instead of centering on the harmlessness of LLM-generated content in most prior studies, this work addresses the imperative need for benchmarking the behavioral safety of LLM agents within diverse environments. We introduce R-Judge, a benchmark crafted to evaluate the proficiency of LLMs in judging and identifying safety risks given agent interaction records. R-Judge comprises 569 records of multi-turn agent interaction, encompassing 27 key risk scenarios among 5 application categories and 10 risk types. It is of high-quality curation with annotated safety labels and risk descriptions. Evaluation of 11 LLMs on R-Judge shows considerable room for enhancing the risk awareness of LLMs: The best-performing model, GPT-4o, achieves 74.42% while no other models significantly exceed the random. Moreover, we reveal that risk awareness in open agent scenarios is a multi-dimensional capability involving knowledge and reasoning, thus challenging for LLMs. With further experiments, we find that fine-tuning on safety judgment significantly improve model performance while straightforward prompting mechanisms fail. R-Judge is publicly available at https://github.com/Lordog/R-Judge.

cs.CL↗

arXiv2024

Measuring Bargaining Abilities of LLMs: A Benchmark and A Buyer-Enhancement Method

Bargaining is an important and unique part of negotiation between humans. As LLM-driven agents learn to negotiate and act like real humans, how to evaluate agents' bargaining abilities remains an open problem. For the first time, we formally described the Bargaining task as an asymmetric incomplete information game, defining the gains of the Buyer and Seller in multiple bargaining processes. It allows us to quantitatively assess an agent's performance in the Bargain task. We collected a real product price dataset, AmazonHistoryPrice, and conducted evaluations of various LLM agents' bargaining abilities. We find that playing a Buyer is much harder than a Seller, and increasing model size can not effectively improve the Buyer's performance. To address the challenge, we propose a novel approach called OG-Narrator that integrates a deterministic Offer Generator to control the price range of Buyer's offers, and an LLM Narrator to create natural language sentences for generated offers. Experimental results show that OG-Narrator improves the buyer's deal rates from 26.67% to 88.88% and brings a ten times multiplication of profits on all baselines, even a model that has not been aligned.

cs.CL↗

arXiv2024

An input-output framework for stability and synchronization analysis of networks of infinite-dimensional linear systems

This paper presents a synchronization criterion for networks of infinite-dimensional linear systems, extending a previous result for finite-dimensional systems. Our result, established in the general framework of input-output relations, requires an additional input-output stability property, compared to the finite-dimensional counterpart. We show that this this property holds for a large class of infinite-dimensional systems including abstract Cauchy problems, parabolic partial differential equations, and time-delay differential equations.

math.OC↗

arXiv2024

Mitigating attribute amplification in counterfactual image generation

Causal generative modelling is gaining interest in medical imaging due to its ability to answer interventional and counterfactual queries. Most work focuses on generating counterfactual images that look plausible, using auxiliary classifiers to enforce effectiveness of simulated interventions. We investigate pitfalls in this approach, discovering the issue of attribute amplification, where unrelated attributes are spuriously affected during interventions, leading to biases across protected characteristics and disease status. We show that attribute amplification is caused by the use of hard labels in the counterfactual training process and propose soft counterfactual fine-tuning to mitigate this issue. Our method substantially reduces the amplification effect while maintaining effectiveness of generated images, demonstrated on a large chest X-ray dataset. Our work makes an important advancement towards more faithful and unbiased causal modelling in medical imaging.

cs.CV↗