Current Vision Language Models (VLMs) remain vulnerable to malicious prompts that induce harmful outputs. Existing safety benchmarks for VLMs primarily rely on automated evaluation methods, but these methods struggle to detect implicit harmful content or produce inaccurate evaluations. Therefore, we found that existing benchmarks have low levels of harmfulness, ambiguous data, and limited diversity in image-text pair combinations. To address these issues, we propose the ELITE benchmark, a high-quality safety evaluation benchmark for VLMs, underpinned by our enhanced evaluation method, the ELITE evaluator. The ELITE evaluator explicitly incorporates a toxicity score to accurately assess harmfulness in multimodal contexts, where VLMs often provide specific, convincing, but unharmful descriptions of images. We filter out ambiguous and low-quality image-text pairs from existing benchmarks using the ELITE evaluator and generate diverse combinations of safe and unsafe image-text pairs. Our experiments demonstrate that the ELITE evaluator achieves superior alignment with human evaluations compared to prior automated methods, and the ELITE benchmark offers enhanced benchmark quality and diversity. By introducing ELITE, we pave the way for safer, more robust VLMs, contributing essential tools for evaluating and mitigating safety risks in real-world applications.

Domain adaptation is an active area of research driven by the growing demand for robust machine learning models that perform well on real-world data. Adversarial learning for deep neural networks (DNNs) has emerged as a promising approach to improving generalization ability, particularly for image classification. In this paper, we implement a specific adversarial learning technique known as Adversarial Discriminative Domain Adaptation (ADDA) and replicate digit classification experiments from the original ADDA paper. We extend their findings by examining a broader range of domain shifts and provide a detailed analysis of in-domain classification accuracy post-ADDA. Our results demonstrate that ADDA significantly improves accuracy across certain domain shifts with minimal impact on in-domain performance. Furthermore, we provide qualitative analysis and propose potential explanations for ADDA's limitations in less successful domain shifts. Code is at https://github.com/eugenechoi2004/COS429_FINAL .

Reinforcement Learning (RL) has been used to finetune Large Language Models (LLMs) using a reward model trained from preference data, to better align with human judgment. The recently introduced direct alignment methods, which are often simpler, more stable, and computationally lighter, can more directly achieve this. However, these approaches cannot optimize arbitrary rewards, and the preference-based ones are not the only rewards of interest for LLMs (eg., unit tests for code generation or textual entailment for summarization, among others). RL-finetuning is usually done with a variation of policy gradient, which calls for on-policy or near-on-policy samples, requiring costly generations. We introduce Contrastive Policy Gradient, or CoPG, a simple and mathematically principled new RL algorithm that can estimate the optimal policy even from off-policy data. It can be seen as an off-policy policy gradient approach that does not rely on important sampling techniques and highlights the importance of using (the right) state baseline. We show this approach to generalize the direct alignment method IPO (identity preference optimization) and classic policy gradient. We experiment with the proposed CoPG on a toy bandit problem to illustrate its properties, as well as for finetuning LLMs on a summarization task, using a learned reward function considered as ground truth for the purpose of the experiments.

Reinforcement Learning from Human Feedback (RLHF) (Christiano et al., 2017) has rapidly become a standard method to align Large Language Models (LLMs). One of the main practical issues that all the prominent existing RLHF methods (offline or online) (Ouyang et al., 2022; Rafailov et al., 2023; Azar et al., 2023; Zhao et al., 2023b; Ahmadian et al., 2024) encounter is that their optimal solution heavily depends on the training task in terms of the distribution used to generate the preference data (behavior policy) (Munos et al., 2023; Azar et al., 2023). This makes the existing RLHF methods prone to out-of-distribution (OOD) tasks (Li et al., 2024; Kirk et al., 2024) where the evaluation distribution is significantly different from that of the behavior policy. Also, whenever the base/SFT models significantly differ from the behavior policy, the dependency of the RLHF solutions on the behavior policy makes the preference dataset and reward model less useful (Gao et al., 2022) as RLHF may undo the SFT/pretraining. To address this challenge, we introduce an alternative approach for aligning LLMs from human preferences based on more principled and robust foundations. Our goal is to find a solution that is robust to the changes in the preference dataset, meaning that changes in the distribution from which the completions are sampled do not affect the final outcome of learning significantly. To achieve this goal, we exploit the concept of self-improving (Huang et al., 2022; Bai et al., 2022) language models. By self-improving LLM we refer to a model capable of enhancing its outputs recursively with each inference iteration.

We present a collection of explicit formulas for the minimum volume of Sasaki-Einstein 5-manifolds. The cone over these 5-manifolds is a toric Calabi-Yau 3-fold. These toric Calabi-Yau 3-folds are associated with an infinite class of 4d N=1 supersymmetric gauge theories, which are realized as worldvolume theories of D3-branes probing the toric Calabi-Yau 3-folds. Under the AdS/CFT correspondence, the minimum volume of the Sasaki-Einstein base is inversely proportional to the central charge of the corresponding 4d N=1 superconformal field theories. The presented formulas for the minimum volume are in terms of geometric invariants of the toric Calabi-Yau 3-folds. These explicit results are derived by implementing machine learning regularization techniques that advance beyond previous applications of machine learning for determining the minimum volume. Moreover, the use of machine learning regularization allows us to present interpretable and explainable formulas for the minimum volume. Our work confirms that, even for extensive sets of toric Calabi-Yau 3-folds, the proposed formulas approximate the minimum volume with remarkable accuracy.

Recent large-scale neural autoregressive sequence models have shown impressive performances on a variety of natural language generation tasks. However, their generated sequences often exhibit degenerate properties such as non-termination, undesirable repetition, and premature termination, when generated with decoding algorithms such as greedy search, beam search, top-k sampling, and nucleus sampling. In this paper, we focus on the problem of non-terminating sequences resulting from an incomplete decoding algorithm. We first define an incomplete probable decoding algorithm which includes greedy search, top-k sampling, and nucleus sampling, beyond the incomplete decoding algorithm originally put forward by Welleck et al. (2020). We then propose a non-monotonic self-terminating language model, which significantly relaxes the constraint of monotonically increasing termination probability in the originally proposed self-terminating language model by Welleck et al. (2020), to address the issue of non-terminating sequences when using incomplete probable decoding algorithms. We prove that our proposed model prevents non-terminating sequences when using not only incomplete probable decoding algorithms but also beam search. Autoregressive neural sequence models (Bengio et al., 2000) have been widely used for various natural language generation tasks such as language modeling (Brown et al., 2020; Chowdhery et al., 2022), machine translation (Bahdanau et al., 2014), and conversational dialogue modeling (Vinyals & Le, 2015). Furthermore, large-scale autoregressive neural sequence models have shown unprecedented ability to generate fluent, human-like texts (Vaswani et al., 2017; Brown et al., 2020). Despite their success, the autoregressive neural sequence models have shown undesirable behaviors: non-termination (Welleck et al., 2020), degenerate repetition (Welleck et al., 2019; Holtzman et al., 2020), and premature termination (Koehn & Knowles, 2017; Stahlberg & Byrne, 2019). In this paper, we focus on how to prevent non-termination when using a given decoding algorithm. Non-termination is the problem that a language model generates infinitely long sequences with a positive probability from our language model when using a given decoding algorithm.