Searcharxiv⌕ Search

arXiv subjects

Bo Liu

Publications and source records attributed to Bo Liu.

At least 55 records · Page 3Linked to original sources
arXiv2026

ZeroPur: Succinct Training-Free Adversarial Purification

Adversarial purification is a kind of defense technique that can defend against various unseen adversarial attacks without modifying the victim classifier. Existing methods often depend on external generative models or cooperation between auxiliary functions and victim classifiers. However, retraining generative models, auxiliary functions, or victim classifiers relies on the domain of the fine-tuned dataset and is computation-consuming. In this work, we suppose that adversarial images are outliers of the natural image manifold, and the purification process can be considered as returning them to this manifold. Following this assumption, we present a simple adversarial purification method without further training to purify adversarial images, called ZeroPur. ZeroPur contains two steps: given an adversarial example, Guided Shift obtains the shifted embedding of the adversarial example by the guidance of its blurred counterparts; after that, Adaptive Projection constructs a directional vector by this shifted embedding to provide momentum, projecting adversarial images onto the manifold adaptively. ZeroPur is independent of external models and requires no retraining of victim classifiers or auxiliary functions, relying solely on victim classifiers themselves to achieve purification. Extensive experiments on three datasets (CIFAR-10, CIFAR-100, and ImageNet-1K) using various classifier architectures (ResNet, WideResNet) demonstrate that our method achieves state-of-the-art robust performance. The source code is publicly available at: https://github.com/erhul/ZeroPur.

cs.CV↗
arXiv2026

Emotion2Skill: Model-Internal Emotion Signals for Adaptive Skill Selection and Evolution

Skill-based LLM agents select reusable procedures from an external library to solve complex tasks, yet their routing decisions rely entirely on text-level signals such as task descriptions, verbal reflections, and experience-derived rules, while the model's own internal representational state remains unobserved. Recent interpretability work has shown that LLMs maintain linear emotion representations that causally influence behavior; however, these representations have been exploited only for post-hoc analysis or direct output steering, and have not been used to inform agent-level decision-making. We propose Emotion2Skill, a framework that extracts LLM-internal emotion vectors and incorporates them into both skill selection and skill evolution. At each decision step, a 27-dimensional emotion state is extracted from the residual stream and mapped to a confidence-gated summary injected into the routing prompt. Beyond online selection, emotion trajectories are analyzed for abrupt internal-state shifts to pinpoint problematic skill invocations, guiding targeted SOP rewriting that replaces the coarse binary outcome signal of prior methods. On WebShop and ALFWorld, Emotion2Skill with Qwen3-8B improves over the Zero-Shot baseline by +26.9% success rate and +25.5% average success respectively, outperforming all baselines on both benchmarks with consistent gains on Qwen3-14B. Co-activation analysis further reveals semantically coherent emotion--skill pairings, confirming that the routing improvements reflect meaningful internal-state signals rather than opaque statistical correlations. These results establish LLM-internal emotion representations as an effective decision-level signal for orchestrating agent skill systems, extending their utility beyond interpretability and output steering. The code is available at https://github.com/BoHan-LIN04/Emotion2Skill.

cs.AI↗
arXiv2026

DTMC-Based Analysis and Scheduling for Periodic Flows with Proactive HARQ

Ultra-Reliable Low-Latency Communication (URLLC) requires strict reliability and latency guarantees for heterogeneous periodic traffic. Proactive HARQ improves resource efficiency through early termination, but slot-level timing effects, particularly delayed feedback, complicate schedulability analysis. This paper presents a discrete-time Markov chain (DTMC)-based framework for periodic flows with proactive HARQ. By expanding the state space, the model captures HARQ round-trip time and other cross-slot timing effects. The framework determines the transmission opportunities required to satisfy heterogeneous reliability and latency constraints and supports offset-based scheduling through a two-stage genetic algorithm. Simulations with industrial URLLC traffic show that the proposed method achieves higher schedulability than reactive HARQ, K-Repetition, and non-guaranteed proactive HARQ, with acceptable computational overhead.

cs.NI↗
arXiv2026

Maglev: Sliding Recurrent Memory

We introduce \ours{}, a recurrent Transformer architecture with fixed-size memory that generalizes sliding-window attention while remaining parallelizable during training. \ours{} consists of two coupled models: a prefiller $Q$, which leverages full attention\footnote{In practice, we use interleaved full and sliding-window attention for $Q$, as this yields stronger performance. The essential requirement is that $Q$ be more expressive than $P$, with access to the full history.} to produce memory targets $m'_t$, and a decoder $P$, which uses only sliding-window attention and recurrent K/V injection to produce decoder memories $m_t$ for next-token prediction. We train \ours{} with a memory consistency loss that aligns $m_t$ with $m'_t$, allowing inference to use $P$ alone. Empirically, \ours{} improves validation loss and downstream pretraining benchmarks over sliding-window and latent recurrent transformer baselines. Moreover, sharing parameters between $P$ and $Q$ reduces parameter memory while preserving most of the gains.

cs.LG↗
arXiv2026

From RLVR to RLSVR: Task Transformation Induces Self-Verifiable Rewards for Open-Ended LLM Self-Improvement

Reinforcement Learning with Verifiable Rewards (RLVR) has driven recent progress in reasoning-oriented large language models (LLMs) by enabling large-scale optimization. However, its applicability remains largely limited to domains such as mathematics and coding, where correctness can be deterministically verifiable. Open-ended tasks instead often rely on human preferences, reward models, or LLM-based judges, introducing evaluation bias, judge capability bottlenecks, and additional inference costs. Drawing on the principle of self-supervised learning, which constructs pretext tasks to derive supervision from the data itself, we propose Reinforcement Learning with Self-Verifiable Rewards (RLSVR), a task-transformation-based training paradigm for extending RLVR to open-ended tasks. RLSVR transforms open-ended tasks into verifiable proxy environments whose internal rules and interaction outcomes automatically generate reward signals. We instantiate RLSVR with SpyRL, a Self-PlaY Reinforcement Learning method inspired by social deduction game Who Is the Spy?. Agents receive asymmetric information, complete the same target task, and vote to identify a designated spy. Because the spy identity is predetermined, voting outcomes provide fully verifiable rewards, while successful identification remains closely related to output quality. Experiments on text summarization, creative writing, and mathematical reasoning show that SpyRL outperforms existing self-improvement methods on non-verifiable tasks and yields consistent gains on verifiable reasoning tasks. These results demonstrate that task transformation can extend scalable RLVR-based self-improvement beyond inherently verifiable domains. Models and code have been released at https://github.com/wangqinsi1/RLSVR/tree/SpyRL.

cs.AI↗
arXiv2026

EntropyMoE: Entropy-Aware Sparse Expert Routing for Tokenizer-Free LLMs

Recent byte-level large language models (LLMs) have made tokenizer-free modeling increasingly competitive by grouping bytes into dynamically sized patches. However, existing byte-patch architectures still apply the same dense feed-forward computation to every patch. This uniform computation cannot adapt model capacity to variations in patch semantics and granularity. We address this limitation with EntropyMoE, a Mixture-of-Experts (MoE) architecture designed for dynamic byte patches. EntropyMoE replaces the dense feed-forward modules in the global patch Transformer with Top-K expert layers. Each dynamic patch serves as the basic unit of expert routing, and its byte coverage determines its contribution to workload accounting. The router selects experts directly from patch entropy, using the same granularity signal that underlies dynamic patch construction to organize sparse computation. Patch entropy and length jointly define the feature space for regulating expert specialization. Experiments show that EntropyMoE achieves the lowest held-out bits-per-byte among matched dense and sparse baselines while maintaining comparable downstream accuracy. These results establish patch entropy as an effective routing coordinate for sparse conditional computation and extend Mixture-of-Experts modeling beyond tokenizer-based representations.

cs.AI↗
arXiv2026

SpecCal: Ambiguity-Aware Candidate Calibration for Infrared Spectrum-Based Molecular Structure Reconstruction

Inferring molecular structures from infrared (IR) spectra is a fundamental yet challenging problem. A key difficulty is that an IR spectrum provides limited structural information: different molecules may share similar functional groups and local vibrational patterns, leading to highly similar spectral responses. Thus, even when an observed spectrum has a unique underlying structure, reconstructing it from the spectrum remains ambiguous. Existing IR-to-molecule models usually generate a ranked set of candidate molecules, but this set is largely determined by the model's learned generation preference and may not fully capture the structures that best satisfy the observed spectral constraints. To address this limitation, we propose SpecCal, a training-free candidate calibration framework for IR-to-molecule prediction. SpecCal operates on the candidate outputs of existing base models and improves the prediction set by re-ranking current candidates while introducing additional structurally plausible alternatives guided by spectral consistency. The framework is plug-and-play and model-agnostic, requiring no parameter updates for integration with diverse base models. Experiments on multiple benchmarks show that SpecCal consistently improves top-k reconstruction at both SMILES and scaffold levels across different base models. Further analyses demonstrate that calibrating candidate sets under spectral ambiguity provides a practical way to improve molecular reconstruction from IR spectra. The code is available at: https://anonymous.4open.science/r/SpecCal-B18A.

cs.AI↗
arXiv2026

Deep Expert Injection for Anchoring Retinal VLMs with Domain-Specific Knowledge

Large Vision Language Models (LVLMs) show immense potential for automated ophthalmic diagnosis. However, their clinical deployment is severely hindered by lacking domain-specific knowledge. In this work, we identify two structural deficiencies hindering reliable medical reasoning: 1) the Perception Gap, where general-purpose visual encoders fail to resolve fine-grained pathological cues (e.g., microaneurysms); and 2) the Reasoning Gap, where sparse visual evidence is progressively overridden by massive language priors in deeper transformer layers, leading to ungrounded hallucinations. To bridge these gaps, we propose EyExIn, a data-efficient framework designed to anchor retinal VLMs with expert knowledge via a Deep Expert Injection mechanism. Our architecture employs an Expert-Aware Dual-Stream encoding strategy that decouples visual representation into a general stream for anatomical context and a specialized expert stream for pathological semantics. To ensure high-fidelity integration, we design a Semantic-Adaptive Gated Fusion module, which dynamically amplifies subtle lesion signals while filtering irrelevant background noise. Furthermore, we introduce Adaptive Deep Expert Injection to embed persistent "Vision Anchors" by integrating fused visual features as residual biases directly into intermediate LLM layers. This mechanism creates a visual shortcut that forces the reasoning stack to remain strictly grounded in visual evidence. Extensive experiments across four benchmarks demonstrate that our model consistently outperforms massive proprietary systems. EyExIn significantly enhances domain-specific knowledge embedding and achieves state-of-the-art precision in ophthalmic visual question answering, advancing the development of trustworthy ophthalmic AI.

cs.CV↗
arXiv2026

EchoBridge: Long-Tail-Aware ECG-Echocardiography Text Alignment for Echocardiography-Derived Cardiac Findings

Standardized echocardiography conclusions provide meaningful supervision for learning ECG representations of echocardiography-derived cardiac findings. Global ECG--text alignment may entangle modality-specific factors, while long-tailed finding distributions provide sparse positive supervision for low-prevalence conditions. We propose EchoBridge with Complementary Shared--Private Projection (CSPP) and Adaptive Prototype Boundary Calibration (APBC). CSPP maps each modality into shared and auxiliary private projections, reduces directional redundancy via within-modality orthogonality, and bidirectionally aligns normalized shared projections. APBC organizes the shared hypersphere with class-specific prototypes, training-frequency-adaptive angular margins, and spherical Riesz repulsion. We evaluate EchoBridge on EchoNext-Mini and independent PKUPH and SHTMU cohorts under four protocols: prompt-based inference without downstream classifier training, in-domain frozen linear probing, target-domain cross-center frozen linear probing, and source-only cross-center transfer, supplemented by finding-specific analyses. EchoBridge improves classifier-free AUROC, AUPRC, and F1 over the strongest baselines by 7.88, 5.61, and 4.54 points, respectively, and achieves the highest point estimates across all in-domain and target-domain probing budgets and both source-only transfer cohorts. Finding-specific analyses show gains for most conditions, including several low-prevalence valvular findings.

cs.LG↗
arXiv2026

Gumbel Distillation for Parallel Text Generation

The slow, sequential nature of autoregressive (AR) language models has driven the adoption of parallel decoding methods. However, these non-AR models often sacrifice generation quality as they struggle to model the complex joint distribution of token sequences. To narrow this performance gap, we introduce Gumbel Distillation, a novel distillation technique that enables parallel decoders to learn this distribution effectively. Our method leverages the Gumbel-Max trick to create a deterministic mapping from a latent Gumbel noise space to the output tokens of a high-performing AR teacher. As a model-agnostic technique, Gumbel Distillation seamlessly integrates with diverse parallel decoding architectures, including MDLM and BD3-LM. Experiments on LM1B and OpenWebText show that Gumbel Distillation substantially improves the generation quality of parallel language models, achieving a 30.0% improvement in MAUVE score and 10.5% in generative perplexity over MDLM trained on OpenWebText dataset. Code available at https://github.com/hxixixh/gumbel-distill.

cs.CL↗
arXiv2026

REAL: Reading Out Transformer Activations for Precise Localization in Language Model Steering

Inference-time steering aims to alter a large language model's (LLM's) responses without changing its parameters, but a central challenge is identifying the internal modules that most strongly govern the target behavior. Existing approaches often rely on simplistic cues or ad hoc heuristics, leading to suboptimal or unintended effects. We introduce REAL, a framework for identifying behavior-relevant modules (attention heads or layers) in Transformer models. For each module, REAL trains a vector-quantized autoencoder (VQ-AE) on its hidden activations and uses a shared, learnable codebook to partition the latent space into behavior-relevant and behavior-irrelevant subspaces. REAL quantifies a module's behavioral relevance by how well its VQ-AE encodings discriminate behavior-aligned from behavior-violating responses via a binary classification metric; this score guides both module selection and steering strength. We evaluate REAL across eight LLMs from the Llama and Qwen families and nine datasets spanning truthfulness enhancement, open-domain QA under knowledge conflicts, and general alignment tasks. REAL enables more effective inference-time interventions, achieving an average relative improvement of 20% (up to 81.5%) over the ITI method on truthfulness steering. In addition, the modules selected by REAL exhibit strong zero-shot generalization in cross-domain truthfulness-steering scenarios.

cs.CL↗
arXiv2026

Multi-Catheter Digitization in Brachytherapy via Few-Shot Synthetic-to-Real Learning and Structure-Aware Tracking

Accurate catheter digitization in CT-guided interstitial brachytherapy is a critical but time-consuming task, especially for complex implant configurations. We developed a data-efficient, physics-guided framework for automated multi-catheter digitization with minimal clinical annotation. The pipeline consists of two stages. First, an implant region-aware network was pretrained on synthetic CT volumes with simulated metallic signatures and then fine-tuned using only 10 clinical cases. Second, a structure-aware reconstruction module combined a direction-constrained 3D Hough transform with synchronous physics-constrained inward tracking to separate adherent catheter trajectories. The method was evaluated by patient-level five-fold cross-validation on 203 treatment fractions from 38 patients. The fine-tuned network achieved an HD95 of 0.853 +/- 0.362 mm. End-to-end evaluation yielded an F1 score of 0.891 +/- 0.178, with shaft and tip errors of 0.334 +/- 0.367 mm and 0.896 +/- 0.680 mm, respectively. In cases with severe catheter adhesion, the tracking F1 score remained 0.843 +/- 0.190. The complete workflow required approximately 11.6 s per case. These results indicate that combining few-shot synthetic-to-real learning with physics-guided structural tracking can provide robust and efficient multi-catheter digitization for time-sensitive clinical workflows.

physics.med-ph↗
arXiv2026

Simple Recipe Works: Vision-Language-Action Models are Natural Continual Learners with Reinforcement Learning

Continual Reinforcement Learning (CRL) for Vision-Language-Action (VLA) models is a promising direction toward self-improving embodied agents that can adapt in openended, evolving environments. However, conventional wisdom from continual learning suggests that naive Sequential Fine-Tuning (Seq. FT) leads to catastrophic forgetting, necessitating complex CRL strategies. In this work, we take a step back and conduct a systematic study of CRL for large pretrained VLAs across diverse lifelong RL benchmarks. We find that, contrary to established belief, simple Seq. FT with low-rank adaptation (LoRA) is remarkably strong: it achieves high plasticity, exhibits little to no forgetting, and retains strong zero-shot generalization, frequently outperforming more sophisticated CRL methods. Through detailed analysis, we show that this robustness arises from a synergy between the large pretrained model, parameter-efficient adaptation, and on-policy RL. Together, these components reshape the stability-plasticity trade-off, making continual adaptation both stable and scalable. Our results position Sequential Fine-Tuning as a powerful method for continual RL with VLAs and provide new insights into lifelong learning in the large model era. Code is available at https://github.com/UT-Austin-RobIn/continual-vla-rl

cs.LG↗
arXiv2026

TeamTR: Trust-Region Fine-Tuning for Multi-Agent LLM Coordination

Multi-agent LLM systems have shown promise for complex reasoning, yet recent evaluations reveal they often underperform single-model baselines. We identify a structural failure mode in sequential fine-tuning of shared-context teams: updating one agent shifts the team's context distribution, and when subsequent updates are evaluated on cached rollouts, this mismatch compounds. We formalize this as the compounding occupancy shift and prove that stale-occupancy evaluation incurs a penalty that scales quadratically with the number of agents. In contrast, intermediate-occupancy evaluation reduces this to linear scaling. We propose TeamTR, a trust-region framework that resamples trajectories after each component update and enforces per-agent divergence control, yielding rigorous per-update and per-stage improvement lower bounds. Experiments show that TeamTR outperforms single-agent and sequential baselines with 7.1% on average, mitigates coordination regressions, and supports plug-and-play component replacement. Code is available at https://github.com/Yydc/TeamTR.

cs.LG↗
arXiv2026

DICE: Entropy-Regularized Equilibrium Selection for Stable Multi-Agent LLM Coordination

Multi-agent large language model (LLM) systems often fail to reliably outperform a single strong model equipped with best-of-N sampling. We argue that a core source of this instability is ill-posed equilibrium selection: current systems specify what information agents share, but not which coordination convention should be selected. We formalize a broad class of such systems as discounted incomplete-information Markov games and show that two common pathologies, oscillation between competing conventions and drift across them, can both induce unstable learning and linear Bayesian regret. To obtain a well-posed target, we introduce the Heterogeneous Quantal Response Equilibrium (HQRE), an entropy-regularized equilibrium concept with agent- and state-dependent temperatures. Under a monotonicity condition, HQRE is unique, admits linearly convergent mirror updates, and yields bounded Bayesian regret; the same condition yields rollout-measurable stability diagnostics. We instantiate this objective in two algorithms: DICE-PC, which coordinates frozen models through prompt-control actions, and DICE-FT, which performs parameter-efficient mirror fine-tuning. Across eleven benchmarks in four domains, DICE improves accuracy-cost trade-offs over strong within-class baselines; on reasoning and planning tasks, DICE-PC improves by 4.3 percentage points on average and DICE-FT by 8.5 points.

cs.LG↗
arXiv2026

Dzyaloshinskii-Moriya Gradients Unlock Topological Dimensional Reduction in Magnetic Hopfions

Three-dimensional magnetic solitons retain topological protection only while their spin field remains continuous. Here we show that chemical inhomogeneity can break this protection in a controlled way, converting a hopfion-like toroidal texture into an effectively two-dimensional skyrmion string. Tilt-dependent Lorentz transmission electron microscopy, electron energy-loss spectroscopy, and micromagnetic simulations of pristine, uniformly Y-doped, and nonuniformly Y-doped disordered TiO2 nanoparticles embedded in a FeCrNiMn host reveal two regimes. Uniform Y doping enriches Ti3+/oxygen-vacancy localization centers and stabilizes closed rings with Hopf invariant QH approximately 1. Nonuniform Y doping forms a Ti4+-rich, vacancy-depleted boundary that creates a sharp q=D/(2A) gradient and a weak-moment leakage channel. This coupled mismatch torque and continuity leakage split the ring, leaving a skyrmion-string remnant and a field-sensitive helicity texture.

cond-mat.other↗
arXiv2026

PERFOPT-Bench: Evaluating Coding Agents on Software Performance Optimization

Coding-agent benchmarks have largely measured whether agents can produce functionally correct patches, but production software also demands measurable speedups on real execution targets. Performance optimization is a distinct agentic task: agents must profile executions, diagnose cross-layer bottlenecks, edit code without breaking correctness, and verify that gains are reproducible rather than measurement artifacts. We introduce PERFOPT-Bench, a benchmark for evaluating this full performance-engineering loop. Each task provides a correct but deliberately suboptimal codebase and asks the agent to improve a target performance metric; scoring requires hidden correctness tests, verified-speedup measurement, and trajectory-level audit. We evaluate 7 agent stacks with different LLMs and agent frameworks on 7 long-horizon optimization tasks. The results show that optimization performance is workload-dependent rather than determined by model identity alone: no single stack dominates, and changing the agent framework can materially change the same LLM's per-task speedup profile. We further find that raw speedup is unsafe as a benchmark score, since some large gains arise from benchmark-specific shortcut exploitation; an exploratory relay pilot suggests that restarting from an externalized optimization summary can recover additional headroom after an initial session stops. The benchmark and our evaluation are available at: https://anonymous.4open.science/r/Dataset-D3CC.

cs.SE↗
arXiv2026

KOAL: Knowledge-Driven Prostate Cancer Grading with Ordinal-Aware Learning

Non-invasive prediction of Gleason Grade Group (GGG) in prostate cancer using multiparametric MRI (mpMRI) is clinically vital for reducing unnecessary biopsies. Existing GGG prediction methods face two major limitations. First, they often overlook non-image information critical for GGG prediction, including age, prostate-specific antigen (PSA), and expert priors embedded in radiology reports. Second, they tend to oversimplify GGG as flat categorical labels, failing to account for its intrinsic hierarchy of primary and secondary Gleason patterns. To this end, we propose a novel Knowledge-Driven Ordinal-Aware Learning (KOAL) framework with three synergistic modules. Specifically, the Clinical-Context Modulation (CCM) module uses clinical variables (e.g., age and PSA) to dynamically modulate discriminative image representations. The Knowledge-Guided Prototype Alignment (KGPA) module leverages an LLM to extract group-specific expert knowledge from training radiology reports and clinical guidelines, producing offline semantic anchors describing grade-specific radiological findings without requiring patient-specific reports at inference. Through prototype contrastive alignment, patient-specific mpMRI representations are matched with these anchors to promote pathology-aligned representation learning. The Hierarchical Ordinal-aware Constraints (HOC) module decouples primary and secondary Gleason pattern prediction and maps their probabilistic outputs to GGG via a Differentiable Bio-logic Mapping Layer (DBML), ensuring pathological grading consistency. Experiments on public PI-CAI and in-house datasets demonstrate that KOAL outperforms state-of-the-art methods. Code is available at: https://github.com/Gother-GZ/KOAL.

cs.CV↗