Abstract: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.
Abstract:Federated learning (FL) enables collaborative training of deep learning models without requiring data to leave local clients, thereby preserving client privacy. The aggregation process on the server plays a critical role in the performance of the resulting FL model. The most commonly used aggregation method is weighted averaging based on the amount of data from each client, which is thought to reflect each client's contribution. However, this method is prone to model bias, as dishonest clients might report inaccurate training data volumes to the server, which is hard to verify. To address this issue, we propose a novel secure \underline{Fed}erated \underline{D}ata q\underline{u}antity-\underline{a}ware weighted averaging method (FedDua). It enables FL servers to accurately predict the amount of training data from each client based on their local model gradients uploaded. Furthermore, it can be seamlessly integrated into any FL algorithms that involve server-side model aggregation. Extensive experiments on three benchmarking datasets demonstrate that FedDua improves the global model performance by an average of 3.17% compared to four popular FL aggregation methods in the presence of inaccurate client data volume declarations.
Abstract:Recent advancements in Text-to-Image (T2I) generation have significantly enhanced the realism and creativity of generated images. However, such powerful generative capabilities pose risks related to the production of inappropriate or harmful content. Existing defense mechanisms, including prompt checkers and post-hoc image checkers, are vulnerable to sophisticated adversarial attacks. In this work, we propose 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. By iteratively optimizing tokens near these boundaries, TCBS-Attack generates semantically coherent adversarial prompts capable of bypassing multiple defensive layers in T2I models. Extensive experiments demonstrate that our method 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 45\% and an ASR-1 of 21\% on jailbreaking full-chain T2I models, significantly surpassing baseline methods.
Abstract:Dynamic scheduling in real-world environments often struggles to adapt to unforeseen disruptions, making traditional static scheduling methods and human-designed heuristics inadequate. This paper introduces an innovative approach that combines Genetic Programming (GP) with a Transformer trained through Reinforcement Learning (GPRT), specifically designed to tackle the complexities of dynamic scheduling scenarios. GPRT leverages the Transformer to refine heuristics generated by GP while also seeding and guiding the evolution of GP. This dual functionality enhances the adaptability and effectiveness of the scheduling heuristics, enabling them to better respond to the dynamic nature of real-world tasks. The efficacy of this integrated approach is demonstrated through a practical application in container terminal truck scheduling, where the GPRT method outperforms traditional GP, standalone Transformer methods, and other state-of-the-art competitors. The key contribution of this research is the development of the GPRT method, which showcases a novel combination of GP and Reinforcement Learning (RL) to produce robust and efficient scheduling solutions. Importantly, GPRT is not limited to container port truck scheduling; it offers a versatile framework applicable to various dynamic scheduling challenges. Its practicality, coupled with its interpretability and ease of modification, makes it a valuable tool for diverse real-world scenarios.
Abstract:Spiking Neural Networks (SNNs) based on Transformers have garnered significant attention due to their superior performance and high energy efficiency. However, the spiking attention modules of most existing Transformer-based SNNs are adapted from those of analog Transformers, failing to fully address the issue of over-allocating attention to irrelevant contexts. To fix this fundamental yet overlooked issue, we propose a Lateral Inhibition-inspired Spiking Transformer (SpiLiFormer). It emulates the brain's lateral inhibition mechanism, guiding the model to enhance attention to relevant tokens while suppressing attention to irrelevant ones. Our model achieves state-of-the-art (SOTA) performance across multiple datasets, including CIFAR-10 (+0.45%), CIFAR-100 (+0.48%), CIFAR10-DVS (+2.70%), N-Caltech101 (+1.94%), and ImageNet-1K (+1.6%). Notably, on the ImageNet-1K dataset, SpiLiFormer (69.9M parameters, 4 time steps, 384 resolution) outperforms E-SpikeFormer (173.0M parameters, 8 time steps, 384 resolution), a SOTA spiking Transformer, by 0.46% using only 39% of the parameters and half the time steps. Our code and training checkpoints will be released upon acceptance.
Abstract:Assessing whether AI-generated images are substantially similar to copyrighted works is a crucial step in resolving copyright disputes. In this paper, we propose CopyJudge, an automated copyright infringement identification framework that leverages large vision-language models (LVLMs) to simulate practical court processes for determining substantial similarity between copyrighted images and those generated by text-to-image diffusion models. Specifically, we employ an abstraction-filtration-comparison test framework with multi-LVLM debate to assess the likelihood of infringement and provide detailed judgment rationales. Based on the judgments, we further introduce a general LVLM-based mitigation strategy that automatically optimizes infringing prompts by avoiding sensitive expressions while preserving the non-infringing content. Besides, our approach can be enhanced by exploring non-infringing noise vectors within the diffusion latent space via reinforcement learning, even without modifying the original prompts. Experimental results show that our identification method achieves comparable state-of-the-art performance, while offering superior generalization and interpretability across various forms of infringement, and that our mitigation method could more effectively mitigate memorization and IP infringement without losing non-infringing expressions.
Abstract:State-of-the-art brain graph analysis methods fail to fully encode the small-world architecture of brain graphs (accompanied by the presence of hubs and functional modules), and therefore lack biological plausibility to some extent. This limitation hinders their ability to accurately represent the brain's structural and functional properties, thereby restricting the effectiveness of machine learning models in tasks such as brain disorder detection. In this work, we propose a novel Biologically Plausible Brain Graph Transformer (BioBGT) that encodes the small-world architecture inherent in brain graphs. Specifically, we present a network entanglement-based node importance encoding technique that captures the structural importance of nodes in global information propagation during brain graph communication, highlighting the biological properties of the brain structure. Furthermore, we introduce a functional module-aware self-attention to preserve the functional segregation and integration characteristics of brain graphs in the learned representations. Experimental results on three benchmark datasets demonstrate that BioBGT outperforms state-of-the-art models, enhancing biologically plausible brain graph representations for various brain graph analytical tasks
Abstract:Federated Learning (FL) is notorious for its vulnerability to Byzantine attacks. Most current Byzantine defenses share a common inductive bias: among all the gradients, the densely distributed ones are more likely to be honest. However, such a bias is a poison to Byzantine robustness due to a newly discovered phenomenon in this paper - gradient skew. We discover that a group of densely distributed honest gradients skew away from the optimal gradient (the average of honest gradients) due to heterogeneous data. This gradient skew phenomenon allows Byzantine gradients to hide within the densely distributed skewed gradients. As a result, Byzantine defenses are confused into believing that Byzantine gradients are honest. Motivated by this observation, we propose a novel skew-aware attack called STRIKE: first, we search for the skewed gradients; then, we construct Byzantine gradients within the skewed gradients. Experiments on three benchmark datasets validate the effectiveness of our attack
Abstract:Robotic manipulation is often challenging due to the long-horizon tasks and the complex object relationships. A common solution is to develop a task and motion planning framework that integrates planning for high-level task and low-level motion. Recently, inspired by the powerful reasoning ability of Large Language Models (LLMs), LLM-based planning approaches have achieved remarkable progress. However, these methods still heavily rely on expert-specific knowledge, often generating invalid plans for unseen and unfamiliar tasks. To address this issue, we propose an innovative language-guided symbolic task planning (LM-SymOpt) framework with optimization. It is the first expert-free planning framework since we combine the world knowledge from LLMs with formal reasoning, resulting in improved generalization capability to new tasks. Specifically, differ to most existing work, our LM-SymOpt employs LLMs to translate natural language instructions into symbolic representations, thereby representing actions as high-level symbols and reducing the search space for planning. Next, after evaluating the action probability of completing the task using LLMs, a weighted random sampling method is introduced to generate candidate plans. Their feasibility is assessed through symbolic reasoning and their cost efficiency is then evaluated using trajectory optimization for selecting the optimal planning. Our experimental results show that LM-SymOpt outperforms existing LLM-based planning approaches.
Abstract:Schema, a form of structured knowledge that promotes transfer learning, is attracting growing attention in both neuroscience and artificial intelligence (AI). Current schema research in neural computation is largely constrained to a single behavioral paradigm and relies heavily on recurrent neural networks (RNNs) which lack the neural plausibility and biological interpretability. To address these limitations, this work first constructs a generalized behavioral paradigm framework for schema learning and introduces three novel cognitive tasks, thus supporting a comprehensive schema exploration. Second, we propose a new model using recurrent spiking neural networks with hierarchical intrinsic excitability modulation (HM-RSNNs). The top level of the model selects excitability properties for task-specific demands, while the bottom level fine-tunes these properties for intra-task problems. Finally, extensive visualization analyses of HM-RSNNs are conducted to showcase their computational advantages, track the intrinsic excitability evolution during schema learning, and examine neural coordination differences across tasks. Biologically inspired lesion studies further uncover task-specific distributions of intrinsic excitability within schemas. Experimental results show that HM-RSNNs significantly outperform RSNN baselines across all tasks and exceed RNNs in three novel cognitive tasks. Additionally, HM-RSNNs offer deeper insights into neural dynamics underlying schema learning.