arXiv subjects

Yaochu Jin

Publications and source records attributed to Yaochu Jin.

At least 37 records · Page 2Linked to original sources

arXiv2026

Token-Level Constraint Boundary Search for Jailbreaking Text-to-Image Models

Text-to-Image (T2I) generation has advanced rapidly in recent years, but they also raise safety concerns due to the potential production of harmful content. In the practical deployments, T2I services typically adopt full-chain defenses that combine a prompt checker, a securely trained generator, and a post-hoc image checker. Jailbreaking such full-chain systems is challenging in the black-box settings because prompt tokens form a discrete combinatorial space and the attack must satisfy multiple coupled constraints under sparse feedback and limited queries. To address these challenges, we propose Token-level Constraint Boundary Search (TCBS)-Attack, a novel query-based black-box jailbreak attack that searches for tokens located near the decision boundaries defined by text and image checkers. TCBS-Attack incorporates decision boundaries as constraint conditions to guide the evolutionary search of token populations, iteratively optimize tokens near these boundaries. Such evolutionary search process reduces the effective search space and improves query efficiency while preserving semantic coherence. Extensive experiments demonstrate that TCBS-Attack consistently outperforms state-of-the-art jailbreak attacks across various T2I models, including securely trained open-source models and commercial online services like DALL-E 3. TCBS-Attack achieves an ASR-4 of 52.5% and an ASR-1 of 22.0% on jailbreaking full-chain T2I models, significantly surpassing baseline methods.

cs.CV↗

arXiv2026

ULC: A Unified and Fine-Grained Controller for Humanoid Loco-Manipulation

Loco-Manipulation for humanoid robots aims to enable robots to integrate mobility with upper-body tracking capabilities. Most existing approaches adopt hierarchical architectures that decompose control into isolated upper-body (manipulation) and lower-body (locomotion) policies. While this decomposition reduces training complexity, it inherently limits coordination between subsystems and contradicts the unified whole-body control exhibited by humans. We demonstrate that a single unified policy can achieve a combination of tracking accuracy, large workspace, and robustness for humanoid loco-manipulation. We propose the Unified Loco-Manipulation Controller (ULC), a single-policy framework that simultaneously tracks root velocity, root height, torso rotation, and dual-arm joint positions in an end-to-end manner, proving the feasibility of unified control without sacrificing performance. We achieve this unified control through key technologies: sequence skill acquisition for progressive learning complexity, residual action modeling for fine-grained control adjustments, command polynomial interpolation for smooth motion transitions, random delay release for robustness to deploy variations, load randomization for generalization to external disturbances, and center-of-gravity tracking for providing explicit policy gradients to maintain stability. We validate our method on the Unitree G1 humanoid robot with 3-DOF (degrees-of-freedom) waist. Compared with strong baselines, ULC shows better tracking performance to disentangled methods and demonstrating larger workspace coverage. The unified dual-arm tracking enables precise manipulation under external loads while maintaining coordinated whole-body control for complex loco-manipulation tasks.

cs.RO↗

arXiv2026

Biomechanical Comparisons Reveal Divergence of Human and Humanoid Gaits

It remains challenging to achieve human-like locomotion in legged robots due to fundamental discrepancies between biological and mechanical structures. Although imitation learning has emerged as a promising approach for generating natural robotic movements, simply replicating joint angle trajectories fails to capture the underlying principles of human motion. This study proposes a Gait Divergence Analysis Framework (GDAF), a unified biomechanical evaluation framework that systematically quantifies kinematic and kinetic discrepancies between humans and bipedal robots. We apply GDAF to systematically compare human and humanoid locomotion across 28 walking speeds. To enable reproducible analysis, we collect and release a speed-continuous humanoid locomotion dataset from a state-of-the-art humanoid controller. We further provide an open-source implementation of GDAF, including analysis, visualization, and MuJoCo-based tools, enabling quantitative, interpretable, and reproducible biomechanical analysis of humanoid locomotion. Results demonstrate that despite visually human-like motion generated by modern humanoid controllers, significant biomechanical divergence persists across speeds. Robots exhibit systematic deviations in gait symmetry, energy distribution, and joint coordination, indicating that substantial room remains for improving the biomechanical fidelity and energetic efficiency of humanoid locomotion. This work provides a quantitative benchmark for evaluating humanoid locomotion and offers data and versatile tools to support the development of more human-like and energetically efficient locomotion controllers. The data and code will be made publicly available upon acceptance of the paper.

cs.RO↗

arXiv2026

Evolutionary Generative Optimization: Towards Fully Data-Driven Evolutionary Optimization via Generative Learning

Recent advances in data-driven evolutionary algorithms (EAs) have demonstrated the potential of leveraging historical data to improve optimization accuracy and adaptability. Despite these advancements, existing methods remain reliant on handcrafted process-level operators. In contrast, Evolutionary Generative Optimization (EvoGO) is a fully data-driven framework designed from the objective level, enabling autonomous learning of the entire search process. EvoGO streamlines the evolutionary optimization process into three stages: data preparation, model training, and population generation. The data preparation stage constructs a pairwise dataset to enrich training diversity without incurring additional evaluation costs. During model training, a tailored generative model learns to transform inferior solutions into superior ones. In the population generation stage, EvoGO replaces traditional reproduction operators with a scalable and parallelizable generative mechanism. Extensive experiments on numerical benchmarks, classical control problems, and high-dimensional robotic tasks demonstrate that EvoGO consistently converges within merely 10 generations and substantially outperforms a wide spectrum of optimization approaches, including traditional EAs, Bayesian optimization, and reinforcement learning based methods. Code is available at: https://github.com/EMI-Group/evogo

cs.NE↗

arXiv2026

OmniMER: Auxiliary-Enhanced LLM Adaptation for Indonesian Multimodal Emotion Recognition

Indonesian, spoken by over 200 million people, remains underserved in multimodal emotion recognition research despite its dominant presence on Southeast Asian social media platforms. We introduce IndoMER, the first multimodal emotion recognition benchmark for Indonesian, comprising 1,944 video segments from 203 speakers with temporally aligned text, audio, and visual annotations across seven emotion categories. The dataset exhibits realistic challenges including cross-modal inconsistency and long-tailed class distributions shaped by Indonesian cultural communication norms. To address these challenges, we propose OmniMER, a multimodal adaptation framework built upon Qwen2.5-Omni that enhances emotion recognition through three auxiliary modality-specific perception tasks: emotion keyword extraction for text, facial expression analysis for video, and prosody analysis for audio. These auxiliary tasks help the model identify emotion-relevant cues in each modality before fusion, reducing reliance on spurious correlations in low-resource settings. Experiments on IndoMER show that OmniMER achieves 0.582 Macro-F1 on sentiment classification and 0.454 on emotion recognition, outperforming the base model by 7.6 and 22.1 absolute points respectively. Cross-lingual evaluation on the Chinese CH-SIMS dataset further demonstrates the generalizability of the proposed framework. The dataset and code are publicly available. https://github.com/yanxm01/INDOMER

cs.LG↗

arXiv2026

IRIS: Implicit Reward-Guided Internal Sifting for Mitigating Multimodal Hallucination

Hallucination remains a fundamental challenge for Multimodal Large Language Models (MLLMs). While Direct Preference Optimization (DPO) is a key alignment framework, existing approaches often rely heavily on costly external evaluators for scoring or rewriting, incurring off-policy learnability gaps and discretization loss. Due to the lack of access to internal states, such feedback overlooks the fine-grained conflicts between different modalities that lead to hallucinations during generation. To address this issue, we propose IRIS (Implicit Reward-Guided Internal Sifting), which leverages continuous implicit rewards in the native log-probability space to preserve full information density and capture internal modal competition. This on-policy paradigm eliminates learnability gaps by utilizing self-generated preference pairs. By sifting these pairs based on multimodal implicit rewards, IRIS ensures that optimization is driven by signals that directly resolve modal conflicts. Extensive experiments demonstrate that IRIS achieves highly competitive performance on key hallucination benchmarks using only 5.7k samples, without requiring any external feedback during preference alignment. These results confirm that IRIS provides an efficient and principled paradigm for mitigating MLLM hallucinations.

cs.LG↗

arXiv2026

Overlooked Safety Vulnerability in LLMs: Malicious Intelligent Optimization Algorithm Request and its Jailbreak

The widespread deployment of large language models (LLMs) has raised growing concerns about their misuse risks and associated safety issues. While prior studies have examined the safety of LLMs in general usage, code generation, and agent-based applications, their vulnerabilities in automated algorithm design remain underexplored. To fill this gap, this study investigates this overlooked safety vulnerability, with a particular focus on intelligent optimization algorithm design, given its prevalent use in complex decision-making scenarios. We introduce MalOptBench, a benchmark consisting of 60 malicious optimization algorithm requests, and propose MOBjailbreak, a jailbreak method tailored for this scenario. Through extensive evaluation of 13 mainstream LLMs including the latest GPT-5 and DeepSeek-V3.1, we reveal that most models remain highly susceptible to such attacks, with an average attack success rate of 83.59% and an average harmfulness score of 4.28 out of 5 on original harmful prompts, and near-complete failure under MOBjailbreak. Furthermore, we assess state-of-the-art plug-and-play defenses that can be applied to closed-source models, and find that they are only marginally effective against MOBjailbreak and prone to exaggerated safety behaviors. These findings highlight the urgent need for stronger alignment techniques to safeguard LLMs against misuse in algorithm design.

cs.CR↗

arXiv2025

ParetoHqD: Fast Offline Multiobjective Alignment of Large Language Models using Pareto High-quality Data

Aligning large language models with multiple human expectations and values is crucial for ensuring that they adequately serve a variety of user needs. To this end, offline multiobjective alignment algorithms such as the Rewards-in-Context algorithm have shown strong performance and efficiency. However, inappropriate preference representations and training with imbalanced reward scores limit the performance of such algorithms. In this work, we introduce ParetoHqD that addresses the above issues by representing human preferences as preference directions in the objective space and regarding data near the Pareto front as "high-quality" data. For each preference, ParetoHqD follows a two-stage supervised fine-tuning process, where each stage uses an individual Pareto high-quality training set that best matches its preference direction. The experimental results have demonstrated the superiority of ParetoHqD over five baselines on two multiobjective alignment tasks.

cs.LG↗

arXiv2025

Self-motion as a structural prior for coherent and robust formation of cognitive maps

Most computational accounts of cognitive maps assume that stability is achieved primarily through sensory anchoring, with self-motion contributing to incremental positional updates only. However, biological spatial representations often remain coherent even when sensory cues degrade or conflict, suggesting that self-motion may play a deeper organizational role. Here, we show that self-motion can act as a structural prior that actively organizes the geometry of learned cognitive maps. We embed a path-integration-based motion prior in a predictive-coding framework, implemented using a capacity-efficient, brain-inspired recurrent mechanism combining spiking dynamics, analog modulation and adaptive thresholds. Across highly aliased, dynamically changing and naturalistic environments, this structural prior consistently stabilizes map formation, improving local topological fidelity, global positional accuracy and next-step prediction under sensory ambiguity. Mechanistic analyses reveal that the motion prior itself encodes geometrically precise trajectories under tight constraints of internal states and generalizes zero-shot to unseen environments, outperforming simpler motion-based constraints. Finally, deployment on a quadrupedal robot demonstrates that motion-derived structural priors enhance online landmark-based navigation under real-world sensory variability. Together, these results reframe self-motion as an organizing scaffold for coherent spatial representations, showing how brain-inspired principles can systematically strengthen spatial intelligence in embodied artificial agents.

q-bio.NC↗

arXiv2025

LacaDM: A Latent Causal Diffusion Model for Multiobjective Reinforcement Learning

Multiobjective reinforcement learning (MORL) poses significant challenges due to the inherent conflicts between objectives and the difficulty of adapting to dynamic environments. Traditional methods often struggle to generalize effectively, particularly in large and complex state-action spaces. To address these limitations, we introduce the Latent Causal Diffusion Model (LacaDM), a novel approach designed to enhance the adaptability of MORL in discrete and continuous environments. Unlike existing methods that primarily address conflicts between objectives, LacaDM learns latent temporal causal relationships between environmental states and policies, enabling efficient knowledge transfer across diverse MORL scenarios. By embedding these causal structures within a diffusion model-based framework, LacaDM achieves a balance between conflicting objectives while maintaining strong generalization capabilities in previously unseen environments. Empirical evaluations on various tasks from the MOGymnasium framework demonstrate that LacaDM consistently outperforms the state-of-art baselines in terms of hypervolume, sparsity, and expected utility maximization, showcasing its effectiveness in complex multiobjective tasks.

cs.LG↗

arXiv2025

When Alignment Fails: Multimodal Adversarial Attacks on Vision-Language-Action Models

Vision-Language-Action models (VLAs) have recently demonstrated remarkable progress in embodied environments, enabling robots to perceive, reason, and act through unified multimodal understanding. Despite their impressive capabilities, the adversarial robustness of these systems remains largely unexplored, especially under realistic multimodal and black-box conditions. Existing studies mainly focus on single-modality perturbations and overlook the cross-modal misalignment that fundamentally affects embodied reasoning and decision-making. In this paper, we introduce VLA-Fool, a comprehensive study of multimodal adversarial robustness in embodied VLA models under both white-box and black-box settings. VLA-Fool unifies three levels of multimodal adversarial attacks: (1) textual perturbations through gradient-based and prompt-based manipulations, (2) visual perturbations via patch and noise distortions, and (3) cross-modal misalignment attacks that intentionally disrupt the semantic correspondence between perception and instruction. We further incorporate a VLA-aware semantic space into linguistic prompts, developing the first automatically crafted and semantically guided prompting framework. Experiments on the LIBERO benchmark using a fine-tuned OpenVLA model reveal that even minor multimodal perturbations can cause significant behavioral deviations, demonstrating the fragility of embodied multimodal alignment.

cs.CV↗

arXiv2025

Robustness of Probabilistic Models to Low-Quality Data: A Multi-Perspective Analysis

A systematic, comparative investigation into the effects of low-quality data reveals a stark spectrum of robustness across modern probabilistic models. We find that autoregressive language models, from token prediction to sequence-to-sequence tasks, are remarkably resilient (for GPT-2, test NLL increases modestly from 2.87 to 3.59 despite 50% token corruption). By contrast, under the same levels of data corruption, class-conditional diffusion models degrade catastrophically (image-label consistency plummets by 56.81% relative to baseline), while classifiers show a moderate impact that diminishes with dataset scale. To explain these discrepancies, we analyze the results through a multi-perspective lens, integrating information theory, PAC learning, and gradient dynamics. These analyses suggest that robustness is heavily influenced by two key principles: the richness of conditioning information, which constrains the learning problem, and the absolute information content of the training data, which allows the signal from correct information to dominate statistical noise.

cs.AI↗

arXiv2025

HCPO: Hierarchical Conductor-Based Policy Optimization in Multi-Agent Reinforcement Learning

In cooperative Multi-Agent Reinforcement Learning (MARL), efficient exploration is crucial for optimizing the performance of joint policy. However, existing methods often update joint policies via independent agent exploration, without coordination among agents, which inherently constrains the expressive capacity and exploration of joint policies. To address this issue, we propose a conductor-based joint policy framework that directly enhances the expressive capacity of joint policies and coordinates exploration. In addition, we develop a Hierarchical Conductor-based Policy Optimization (HCPO) algorithm that instructs policy updates for the conductor and agents in a direction aligned with performance improvement. A rigorous theoretical guarantee further establishes the monotonicity of the joint policy optimization process. By deploying local conductors, HCPO retains centralized training benefits while eliminating inter-agent communication during execution. Finally, we evaluate HCPO on three challenging benchmarks: StarCraftII Multi-agent Challenge, Multi-agent MuJoCo, and Multi-agent Particle Environment. The results indicate that HCPO outperforms competitive MARL baselines regarding cooperative efficiency and stability.

cs.LG↗

arXiv2025

Multidomain Evolutionary Optimization on Combinatorial Problems in Complex Networks

Knowledge transfer-based evolutionary optimization has garnered significant attention, such as in multi-task evolutionary optimization (MTEO), which aims to solve complex problems by simultaneously optimizing multiple tasks. While this emerging paradigm has been primarily focusing on task similarity, there remains a hugely untapped potential in harnessing the shared characteristics between different domains. For example, real-world complex systems usually share the same characteristics, such as the power-law rule, small-world property and community structure, thus making it possible to transfer solutions optimized in one system to another to facilitate the optimization. Drawing inspiration from this observation of shared characteristics within complex systems, we present a novel framework, multi-domain evolutionary optimization (MDEO). First, we propose a community-level measurement of graph similarity to manage the knowledge transfer among domains. Furthermore, we develop a graph learning-based network alignment model that serves as the conduit for effectively transferring solutions between different domains. Moreover, we devise a self-adaptive mechanism to determine the number of transferred solutions from different domains, and introduce a knowledge-guided mutation mechanism that adaptively redefines mutation candidates to facilitate the utilization of knowledge from other domains. To evaluate its performance, we use a challenging combinatorial problem known as adversarial link perturbation as the primary illustrative optimization task. Experiments on multiple real-world networks of different domains demonstrate the superiority of the proposed framework in efficacy compared to classical evolutionary optimization.

cs.NE↗

arXiv2025

Mitigating Visual Hallucinations via Semantic Curriculum Preference Optimization in MLLMs

Multimodal Large Language Models (MLLMs) have significantly improved the performance of various tasks, but continue to suffer from visual hallucinations, a critical issue where generated responses contradict visual evidence. While Direct Preference Optimization(DPO) is widely used for alignment, its application to MLLMs often fails to capture fine-grained semantic differences and encourages shortcut learning. To address these challenges, we propose Semantic Curriculum Preference Optimization (SCPO), a novel framework for MLLM alignment. SCPO employs a progressive, easy-to-hard curriculum built upon our Semantic Curriculum Preference Pairs dataset, which provides fine-grained semantic contrasts sorted by difficulty. This curriculum is trained with a dynamic reference model and a novel symmetric, bidirectional objective to facilitate simultaneous learning from both textual and visual preferences. To our knowledge, SCPO is the first framework to unify semantics, symmetry, and curriculum for MLLMs alignment, effectively mitigating visual hallucinations. Extensive experiments on LLaVA models across various scales and versions validate that SCPO demonstrates superior performance compared to baseline models on multiple hallucination benchmarks, reducing the hallucination rate by up to 62.9%. Moreover, evaluations on generalized benchmarks show that SCPO improves factuality while preserving general capabilities, with its performance remaining stable across general vision-language benchmarks.

cs.CV↗

arXiv2025

IP$^{2}$-RSNN: Bi-level Intrinsic Plasticity Enables Learning-to-learn in Recurrent Spiking Neural Networks

Learning-to-learn (L2L), defined as progressively faster learning across similar tasks, is fundamental to both neuroscience and artificial intelligence. However, its neural basis remains elusive, as most studies emphasize neural population dynamics induced by synaptic plasticity while overlooking adaptations driven by intrinsic neuronal plasticity, which point-neuron models cannot capture. To address the above issue, we develop a recurrent spiking neural network with bi-level intrinsic plasticity (IP$^{2}$-RSNN). First, based on task demands, a slow meta-intrinsic plasticity determines which intrinsic neuronal properties are learnable, which is preserved throughout subsequent task learning once configured. Second, a fast intrinsic plasticity fine-tunes those learnable properties within each task. Our results indicate that the proposed bi-level intrinsic plasticity plays a critical role in enabling L2L in RSNNs and show that IP$^{2}$-RSNNs outperform point-neuron recurrent neural networks and self-attention models. Furthermore, our analysis of multi-scale neural dynamics reveals that the bi-level intrinsic plasticity is essential to task-type-specific adaptations at both the neuronal and network levels during L2L, while such adaptations cannot be captured by point-neuron models. Our results suggest that intrinsic plasticity provides significant computational advantages in L2L, shedding light on the design of brain-inspired deep learning models and algorithms.

cs.NE↗

arXiv2025

Enhancing the Effectiveness and Durability of Backdoor Attacks in Federated Learning through Maximizing Task Distinction

Federated learning allows multiple participants to collaboratively train a central model without sharing their private data. However, this distributed nature also exposes new attack surfaces. In particular, backdoor attacks allow attackers to implant malicious behaviors into the global model while maintaining high accuracy on benign inputs. Existing attacks usually rely on fixed patterns or adversarial perturbations as triggers, which tightly couple the main and backdoor tasks. This coupling makes them vulnerable to dilution by honest updates and limits their persistence under federated defenses. In this work, we propose an approach to decouple the backdoor task from the main task by dynamically optimizing the backdoor trigger within a min-max framework. The inner layer maximizes the performance gap between poisoned and benign samples, ensuring that the contributions of benign users have minimal impact on the backdoor. The outer process injects the adaptive triggers into the local model. We evaluate our method on both computer vision and natural language tasks, and compare it with six backdoor attack methods under six defense algorithms. Experimental results show that our method achieves good attack performance and can be easily integrated into existing backdoor attack techniques.

cs.LG↗

arXiv2025

DD-Ranking: Rethinking the Evaluation of Dataset Distillation

In recent years, dataset distillation has provided a reliable solution for data compression, where models trained on the resulting smaller synthetic datasets achieve performance comparable to those trained on the original datasets. To further improve the performance of synthetic datasets, various training pipelines and optimization objectives have been proposed, greatly advancing the field of dataset distillation. Recent decoupled dataset distillation methods introduce soft labels and stronger data augmentation during the post-evaluation phase and scale dataset distillation up to larger datasets (e.g., ImageNet-1K). However, this raises a question: Is accuracy still a reliable metric to fairly evaluate dataset distillation methods? Our empirical findings suggest that the performance improvements of these methods often stem from additional techniques rather than the inherent quality of the images themselves, with even randomly sampled images achieving superior results. Such misaligned evaluation settings severely hinder the development of DD. Therefore, we propose DD-Ranking, a unified evaluation framework, along with new general evaluation metrics to uncover the true performance improvements achieved by different methods. By refocusing on the actual information enhancement of distilled datasets, DD-Ranking provides a more comprehensive and fair evaluation standard for future research advancements.

cs.CV↗