This paper presents a simple, effective, and cost-efficient strategy, named ModelSwitch, to improve LLM performance by scaling test-time compute. ModelSwitch builds upon the repeated-sampling-then-voting framework, with a novel twist: incorporating multiple models, even weaker ones, to leverage their complementary strengths that potentially arise from diverse training data and paradigms. By using sample consistency as a signal, our strategy dynamically switches between models. Theoretical analysis highlights the efficiency and performance advantages of our strategy. Extensive experiments on seven datasets demonstrate that our strategy not only outperforms self-consistency and state-of-the-art multi-agent debate approaches, but also significantly reduces inference costs. Additionally, our strategy requires only a few comparable LLMs to achieve optimal performance and can be extended with verification methods, demonstrating the potential of leveraging multiple LLMs in the generation-verification paradigm.

Balancing performance and efficiency is a central challenge in large language model (LLM) advancement. GPT-5 addresses this with test-time routing, dynamically assigning queries to either an efficient or a high-capacity model during inference. In this work, we present Avengers-Pro, a test-time routing framework that ensembles LLMs of varying capacities and efficiencies, providing a unified solution for all performance-efficiency tradeoffs. The Avengers-Pro embeds and clusters incoming queries, then routes each to the most suitable model based on a performance-efficiency score. Across 6 challenging benchmarks and 8 leading models—including GPT-5-medium, Gemini-2.5-pro, and Claude-opus-4.1—Avengers-Pro achieves state-of-the-art results: by varying a performance-efficiency trade-off parameter, it can surpass the strongest single model (GPT-5-medium) by +7% in average accuracy. Moreover, it can match the average accuracy of the strongest single model at 27% lower cost, and reach ∼ 90% of that performance at 63% lower cost. Last but not least, it achieves a Pareto frontier, consistently yielding the highest accuracy for any given cost, and the lowest cost for any given accuracy, among all single models. Code is available at https://github.com/ZhangYiqun018/AvengersPro.

Huaisheng Zhu, Siyuan Xu, Hangfan Zhang, Teng Xiao, Zhimeng Guo, Shijie Zhou, Shuyue Hu, Vasant G. Honavar. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

A common and effective strategy for humans to improve an unsatisfactory outcome in daily life is to find a cause of this outcome and correct the cause. In this paper, we investigate whether this human improvement strategy can be applied to improving reinforcement learning from human feedback (RLHF) for alignment of language models (LMs). In particular, it is observed in the literature that LMs tuned by RLHF can still output unsatisfactory responses. This paper proposes a method to improve the unsatisfactory responses by correcting their causes. Our method has two parts. The first part proposes a post-hoc explanation method to explain why an unsatisfactory response is generated to a prompt by identifying the training data that lead to this response. We formulate this problem as a constrained combinatorial optimization problem where the objective is to find a set of training data closest to this prompt-response pair in a feature representation space, and the constraint is that the prompt-response pair can be decomposed as a convex combination of this set of training data in the feature space. We propose an efficient iterative data selection algorithm to solve this problem. The second part proposes an unlearning method that improves unsatisfactory responses to some prompts by unlearning the training data that lead to these unsatisfactory responses and, meanwhile, does not significantly degrade satisfactory responses to other prompts. Experimental results demonstrate that our algorithm can improve RLHF.

Diffusion models have established themselves as leading techniques for image generation. However, their reliance on an iterative denoising process results in slow sampling speeds, which limits their applicability to interactive and creative applications. An approach to overcoming this limitation involves distilling multistep diffusion models into efficient one-step generators. However, existing distillation methods typically suffer performance degradation or require complex iterative training procedures which increase their complexity and computational cost. In this paper, we propose Contrastive Energy Distillation (CED), a simple yet effective approach to distill multistep diffusion models into effective one-step generators. Our key innovation is the introduction of an unnormalized joint energy-based model (EBM) that represents the generator and an auxiliary score model. CED optimizes a Noise Contrastive Estimation (NCE) objective to efficiently transfers knowledge from a multistep teacher diffusion model without additional modules or iterative training complexity. We further show that CED implicitly optimizes the KL divergence between the distributions modeled by the multistep diffusion model and the one-step generator. We present results of experiments which demonstrate that CED achieves competitive performance with the representative baselines for distilling multistep diffusion models while maintaining excellent memory efficiency.

Hangfan Zhang, Zhimeng Guo, Huaisheng Zhu, Bochuan Cao, Lu Lin, Jinyuan Jia, Jinghui Chen, Dinghao Wu. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Graph Contrastive Learning (GCL) has attracted considerable interest due to its impressive node representation learning capability. Despite the wide application of GCL techniques, little attention has been paid to the security of GCL. In this paper, we systematically study the vulnerability of GCL in the presence of malicious backdoor adversaries. In particular, we propose *GCBA*, the first backdoor attack for graph contrastive learning. GCBA incorporates three attacks: poisoning, crafting, and natural backdoor, each targeting one stage of the GCL pipeline. We formulate our attacks as optimization problems and solve them with a novel discrete optimization technique to overcome the discrete nature of graph-structured data. By extensively evaluating GCBA on multiple datasets and GCL methods, we show that our attack can achieve high attack success rates while preserving stealthiness. We further consider potential countermeasures to our attack and conclude that existing defenses are insufficient to mitigate GCBA. We show that as a complex paradigm involving data and model republishing, GCL is vulnerable to backdoor attacks, and specifically designed defenses are needed to mitigate the backdoor attacks on GCL.

Federated Learning (FL) is a distributed machine learning paradigm that allows multiple clients to train a global model collaboratively without sharing their local training data. Due to its distributed nature, many studies have shown that it is vulnerable to backdoor attacks. However, existing studies usually used a predetermined, fixed backdoor trigger or optimized it based solely on the local data and model without considering the global training dynamics. This leads to sub-optimal and less durable attack effectiveness, i.e., their attack success rate is low when the attack budget is limited and decreases quickly if the attacker can no longer perform attacks anymore. To address these limitations, we propose A3FL, a new backdoor attack which adversarially adapts the backdoor trigger to make it less likely to be removed by the global training dynamics. Our key intuition is that the difference between the global model and the local model in FL makes the local-optimized trigger much less effective when transferred to the global model. We solve this by optimizing the trigger to even survive the worst-case scenario where the global model was trained to directly unlearn the trigger. Extensive experiments on benchmark datasets are conducted for twelve existing defenses to comprehensively evaluate the effectiveness of our A3FL. Our code is available at https://github.com/hfzhang31/A3FL.

Language processors, such as compilers and interpreters, are indispensable in building modern software. Errors in language processors can lead to severe consequences, like incorrect functionalities or even malicious attacks. However, it is not trivial to automatically test language processors to find bugs. Existing testing methods (or fuzzers) either fail to generate high-quality (i.e., semantically correct) test cases, or only support limited programming languages.In this paper, we propose POLYGLOT, a generic fuzzing framework that generates high-quality test cases for exploring processors of different programming languages. To achieve the generic applicability, POLYGLOT neutralizes the difference in syntax and semantics of programming languages with a uniform intermediate representation (IR). To improve the language validity, POLYGLOT performs constrained mutation and semantic validation to preserve syntactic correctness and fix semantic errors. We have applied POLYGLOT on 21 popular language processors of 9 programming languages, and identified 173 new bugs, 113 of which are fixed with 18 CVEs assigned. Our experiments show that POLYGLOT can support a wide range of programming languages, and outperforms existing fuzzers with up to 30× improvement in code coverage.

Fuzzing is an increasingly popular technique for verifying software functionalities and finding security vulnerabilities. However, current mutation-based fuzzers cannot effectively test database management systems (DBMSs), which strictly check inputs for valid syntax and semantics. Generation-based testing can guarantee the syntax correctness of the inputs, but it does not utilize any feedback, like code coverage, to guide the path exploration.