A BSTRACT In this paper, we aim to build an adversarially robust zero-shot image classifier. We ground our work on CLIP, a vision-language pre-trained encoder model that can perform zero-shot classification by matching an image with text prompts "a photo of a < class-name> .". Purification is the path we choose since it does not require adversarial training on specific attack types and thus can cope with any foreseen attacks. We then formulate purification risk as the KL divergence between the joint distributions of the purification process of denoising the adversarial samples and the attack process of adding perturbations to benign samples, through bidirectional Stochastic Differential Equations (SDEs). The final derived results inspire us to explore purification in the multi-modal latent space of CLIP . We propose two variants for our CLIPure approach: CLIPure-Diff which models the likelihood of images' latent vectors with the DiffusionPrior module in DaLLE-2 (modeling the generation process of CLIP's latent vectors), and CLIPure-Cos which models the likelihood with the cosine similarity between the embeddings of an image and "a photo of a.". As far as we know, CLIPure is the first purification method in multi-modal latent space and CLIPure-Cos is the first purification method that is not based on generative models, which substantially improves defense efficiency. We conducted extensive experiments on CIFAR-10, ImageNet, and 13 datasets that previous CLIP-based defense methods used for evaluating zero-shot classification robustness. Among them, CLIP (Radford et al., 2021) is an example that is popular, effective, and efficient. CLIP performs zero-shot classification by forming text prompts "a photo of a < class-name> ." of all the candidate categories, and selecting the class with the highest similarity with the image embedding. Despite its efficacy, when facing adversarial attacks, its accuracy can drop to zero, similarly vulnerable to other neural classifiers. Existing methods to enhance adversarial robustness follow two primary paths: adversarial training and purification. Adversarial Training (A T) (Madry et al., 2017; Rebuffi et al., 2021; Wang et al., 2023) incorporates adversarial examples into model training to boost robustness. It often achieves corresponding authors 1 arXiv:2502.18176v2 FARE (Schlarmann et al., 2024) and TeCoA (Mao et al., 2022) are two A T approaches integrated with CLIP, which enhance CLIP's zero-shot classification robustness while harming clean accuracy significantly and do not generalize to other types of attacks.

Large Language Models (LLMs) have proven immensely beneficial in education by capturing vast amounts of literature-based information, allowing them to generate context without relying on external sources. In this paper, we propose a generative AI-powered GATE question-answering framework (GATE stands for Graduate Aptitude Test in Engineering) that leverages LLMs to explain GATE solutions and support students in their exam preparation. We conducted extensive benchmarking to select the optimal embedding model and LLM, evaluating our framework based on criteria such as latency, faithfulness, and relevance, with additional validation through human evaluation. Our chatbot integrates state-of-the-art embedding models and LLMs to deliver accurate, context-aware responses. Through rigorous experimentation, we identified configurations that balance performance and computational efficiency, ensuring a reliable chatbot to serve students' needs. Additionally, we discuss the challenges faced in data processing and modeling and implemented solutions. Our work explores the application of Retrieval-Augmented Generation (RAG) for GATE Q/A explanation tasks, and our findings demonstrate significant improvements in retrieval accuracy and response quality. This research offers practical insights for developing effective AI-driven educational tools while highlighting areas for future enhancement in usability and scalability.

Universal healthcare access is critically needed, especially in resource-limited settings. Large Language Models (LLMs) offer promise for democratizing healthcare with advanced diagnostics, but their reliability requires thorough evaluation, especially in trust-dependent environments. This study assesses LLMs' diagnostic reliability focusing on consistency, manipulation resilience, and contextual integration, crucial for safe and ethical use in universal healthcare. We evaluated leading LLMs using 52 patient cases, expanded into variants with demographic changes, symptom rewordings, and exam modifications, while keeping core diagnoses constant. Manipulation susceptibility was tested by inserting misleading narratives and irrelevant details. Contextual awareness was rvaluated by comparing diagnoses with and without patient history. We analyzed diagnostic change rates and response patterns across manipulations. LLMs showed perfect diagnostic consistency for identical data but significant manipulation susceptibility. Gemini had a 40% diagnosis change rate and ChatGPT 30% with irrelevant details. ChatGPT had a higher context influence rate (77.8% vs. Gemini's 55.6%), but both showed limited nuanced contextual integration, exhibiting anchoring bias by prioritizing salient data over context. LLMs' vulnerability to manipulation and limited contextual awareness pose challenges in clinical use. Unlike clinicians, they may overstate diagnostic certainty without validation. Safeguards and domain-specific designs are crucial for reliable healthcare applications. Broad clinical use without oversight is premature and risky. LLMs can enhance diagnostics with responsible use, but future research is needed to improve manipulation resistance and contextual understanding for safe healthcare democratization.

Retrieval-augmented generation (RAG) encounters challenges when addressing complex queries, particularly multi-hop questions. While several methods tackle multi-hop queries by iteratively generating internal queries and retrieving external documents, these approaches are computationally expensive. In this paper, we identify a three-stage information processing pattern in LLMs during layer-by-layer reasoning, consisting of extraction, processing, and subsequent extraction steps. This observation suggests that the representations in intermediate layers contain richer information compared to those in other layers. Building on this insight, we propose Layer-wise RAG (L-RAG). Unlike prior methods that focus on generating new internal queries, L-RAG leverages intermediate representations from the middle layers, which capture next-hop information, to retrieve external knowledge. L-RAG achieves performance comparable to multi-step approaches while maintaining inference overhead similar to that of standard RAG. Experimental results show that L-RAG outperforms existing RAG methods on open-domain multi-hop question-answering datasets, including MuSiQue, HotpotQA, and 2WikiMultiHopQA. The code is available in https://anonymous.4open.science/r/L-RAG-ADD5/

Large Language Models (LLMs) face significant challenges in maintaining privacy, ethics, and compliance, when sensitive or obsolete data must be selectively removed. Retraining these models from scratch is computationally infeasible, necessitating efficient alternatives. As part of the SemEval 2025 Task 4, this work focuses on the application of selective unlearning in LLMs to address this challenge. In this paper, we present our experiments and findings, primarily leveraging global weight modification to achieve an equilibrium between effectiveness of unlearning, knowledge retention, and target model's post-unlearning utility. We also detail the task-specific evaluation mechanism, results, and challenges. Our algorithms have achieved an aggregate score of 0.409 and 0.389 on the test set for 7B and 1B target models, respectively, demonstrating promising results in verifiable LLM unlearning.

Evolution, the engine behind the survival and growth of life on Earth, operates through the population-based process of reproduction. Inspired by this principle, this paper formally defines a newly emerging problem -- the population-based evolution of large language models (LLMs) -- and introduces a novel framework. Starting with a population of parent LLMs, our framework enables the population to evolve through four key operations: (i) crossover, merging the weights of different parents to create offspring LLMs, (ii) mutation, introducing small, random changes to model weights to foster diversity, (iii) selection, prioritizing high-performing models, and (iv) succession, transferring the learned experience from parent to offspring LLMs. With only 200 samples per new task, the LLM population evolves rapidly to adapt to the task at hand, without any gradients. Experiments on 12 datasets show that our framework consistently outperforms existing multi-LLM merging and adaptation methods, achieving accuracy gains of up to 54.8% over the best LLM in the initial population. Moreover, our framework allows for the evolution of LLMs across multiple new tasks simultaneously, scaling effectively with populations of up to 40 LLMs, and even zero-shot generalization to unseen held-out tasks. We have open-sourced the code on GitHub and released the weights of 10 parent LLMs, fine-tuned from gemma-2-2b-it, on HuggingFace$, enabling reproduction of our proposed framework using just a single 4090 GPU with 24GB memory, without any performance degradation.

Deploying large scale language models on edge devices faces inherent challenges such as high computational demands, energy consumption, and potential data privacy risks. This paper introduces the Shakti Small Language Models (SLMs) Shakti-100M, Shakti-250M, and Shakti-500M which target these constraints headon. By combining efficient architectures, quantization techniques, and responsible AI principles, the Shakti series enables on-device intelligence for smartphones, smart appliances, IoT systems, and beyond. We provide comprehensive insights into their design philosophy, training pipelines, and benchmark performance on both general tasks (e.g., MMLU, Hellaswag) and specialized domains (healthcare, finance, and legal). Our findings illustrate that compact models, when carefully engineered and fine-tuned, can meet and often exceed expectations in real-world edge-AI scenarios.

Large Language Models (LLMs) have been observed to process non-human-readable text sequences, such as jailbreak prompts, often viewed as a bug for aligned LLMs. In this work, we present a systematic investigation challenging this perception, demonstrating that unnatural languages - strings that appear incomprehensible to humans but maintain semantic meanings for LLMs - contain latent features usable by models. Notably, unnatural languages possess latent features that can be generalized across different models and tasks during inference. Furthermore, models fine-tuned on unnatural versions of instruction datasets perform on-par with those trained on natural language, achieving 49.71 win rates in Length-controlled AlpacaEval 2.0 in average across various base models. In addition, through comprehensive analysis, we demonstrate that LLMs process unnatural languages by filtering noise and inferring contextual meaning from filtered words.

Large Language Models (LLMs) have demonstrated strong reasoning capabilities, but their safety under adversarial conditions remains a challenge. This study examines the impact of output length on the robustness of DeepSeek-R1, particularly in Forced Thinking scenarios. We analyze responses across various adversarial prompts and find that while longer outputs can improve safety through self-correction, certain attack types exploit extended generations. Our findings suggest that output length should be dynamically controlled to balance reasoning effectiveness and security. We propose reinforcement learning-based policy adjustments and adaptive token length regulation to enhance LLM safety.

The recent wave of audio foundation models (FMs) could provide new capabilities for conversational modeling. However, there have been limited efforts to evaluate these audio FMs comprehensively on their ability to have natural and interactive conversations. To engage in meaningful conversation with the end user, we would want the FMs to additionally perform a fluent succession of turns without too much overlapping speech or long stretches of silence. Inspired by this, we ask whether the recently proposed audio FMs can understand, predict, and perform turn-taking events? To answer this, we propose a novel evaluation protocol that can assess spoken dialog system's turn-taking capabilities using a supervised model as a judge that has been trained to predict turn-taking events in human-human conversations. Using this protocol, we present the first comprehensive user study that evaluates existing spoken dialogue systems on their ability to perform turn-taking events and reveal many interesting insights, such as they sometimes do not understand when to speak up, can interrupt too aggressively and rarely backchannel. We further evaluate multiple open-source and proprietary audio FMs accessible through APIs on carefully curated test benchmarks from Switchboard to measure their ability to understand and predict turn-taking events and identify significant room for improvement. We will open source our evaluation platform to promote the development of advanced conversational AI systems.