Modern AI assistants have made significant progress in natural language understanding and API/tool integration, with emerging efforts to incorporate diverse interfaces (such as Web interfaces) for enhanced scalability and functionality. However, current approaches that heavily rely on repeated LLM-driven HTML parsing are computationally expensive and error-prone, particularly when handling dynamic web interfaces and multi-step tasks. To overcome these challenges, we introduce PAFFA (Premeditated Actions For Fast Agents), a framework designed to enhance web interaction capabilities through an Action API Library of reusable, verified browser interaction functions. By pre-computing interaction patterns and employing two core methodologies - "Dist-Map" for task-agnostic element distillation and "Unravel" for incremental page-wise exploration - PAFFA reduces inference calls by 87% while maintaining robust performance even as website structures evolve. This framework accelerates multi-page task execution and offers a scalable solution to advance autonomous web agent research.
While Large Language Models (LLMs) have shown remarkable capabilities in natural language understanding and generation, their performance often lags in lower-resource, non-English languages due to biases in the training data. In this work, we explore strategies for adapting the primarily English LLMs (Llama-2 and Llama-3) to Dutch, a language spoken by 30 million people worldwide yet often underrepresented in LLM development. We collect 104GB of Dutch text ($32$B tokens) from various sources to first apply continued pretraining using low-rank adaptation (LoRA), complemented with Dutch posttraining strategies provided by prior work. For Llama-2, we consider using (i) the tokenizer of the original model, and (ii) training a new, Dutch-specific tokenizer combined with embedding reinitialization. We evaluate our adapted models, ChocoLlama-2, both on standard benchmarks and a novel Dutch benchmark, ChocoLlama-Bench. Our results demonstrate that LoRA can effectively scale for language adaptation, and that tokenizer modification with careful weight reinitialization can improve performance. Notably, Llama-3 was released during the course of this project and, upon evaluation, demonstrated superior Dutch capabilities compared to our Dutch-adapted versions of Llama-2. We hence apply the same adaptation technique to Llama-3, using its original tokenizer. While our adaptation methods enhanced Llama-2's Dutch capabilities, we found limited gains when applying the same techniques to Llama-3. This suggests that for ever improving, multilingual foundation models, language adaptation techniques may benefit more from focusing on language-specific posttraining rather than on continued pretraining. We hope this work contributes to the broader understanding of adapting LLMs to lower-resource languages, and to the development of Dutch LLMs in particular.
Existing Hardware Trojans (HT) detection methods face several critical limitations: logic testing struggles with scalability and coverage for large designs, side-channel analysis requires golden reference chips, and formal verification methods suffer from state-space explosion. The emergence of Large Language Models (LLMs) offers a promising new direction for HT detection by leveraging their natural language understanding and reasoning capabilities. For the first time, this paper explores the potential of general-purpose LLMs in detecting various HTs inserted in Register Transfer Level (RTL) designs, including SRAM, AES, and UART modules. We propose a novel tool for this goal that systematically assesses state-of-the-art LLMs (GPT-4o, Gemini 1.5 pro, and Llama 3.1) in detecting HTs without prior fine-tuning. To address potential training data bias, the tool implements perturbation techniques, i.e., variable name obfuscation, and design restructuring, that make the cases more sophisticated for the used LLMs. Our experimental evaluation demonstrates perfect detection rates by GPT-4o and Gemini 1.5 pro in baseline scenarios (100%/100% precision/recall), with both models achieving better trigger line coverage (TLC: 0.82-0.98) than payload line coverage (PLC: 0.32-0.46). Under code perturbation, while Gemini 1.5 pro maintains perfect detection performance (100%/100%), GPT-4o (100%/85.7%) and Llama 3.1 (66.7%/85.7%) show some degradation in detection rates, and all models experience decreased accuracy in localizing both triggers and payloads. This paper validates the potential of LLM approaches for hardware security applications, highlighting areas for future improvement.
The increasing sizes of large language models (LLMs) result in significant computational overhead and memory usage when adapting these models to specific tasks or domains. Various parameter-efficient fine-tuning (PEFT) methods have been devised to mitigate these challenges by training a small set of parameters for the task-specific updates of the model weights. Among PEFT methods, LoRA stands out for its simplicity and efficiency, inspiring the development of a series of variants. However, LoRA and its successors disregard the knowledge that is noisy or irrelevant to the targeted task, detrimentally impacting model performance and leading to suboptimality. To address this limitation, we introduce Knowledge-aware Singular-value Adaptation (KaSA), a PEFT method that leverages singular value decomposition (SVD) with knowledge-aware singular values to dynamically activate knowledge based on its relevance to the task at hand. We conduct extensive experiments across a range of LLMs on tasks spanning natural language understanding (NLU), generation (NLG), instruction following, and commonsense reasoning. The experimental results demonstrate that KaSA consistently outperforms FFT and 14 popular PEFT baselines across 16 benchmarks and 4 synthetic datasets, underscoring our method's efficacy and adaptability. The source code of our method is available at https://github.com/juyongjiang/KaSA.
Large Language Models (LLMs) have transformed numerous domains by providing advanced capabilities in natural language understanding, generation, and reasoning. Despite their groundbreaking applications across industries such as research, healthcare, and creative media, their rapid adoption raises critical concerns regarding sustainability. This survey paper comprehensively examines the environmental, economic, and computational challenges associated with LLMs, focusing on energy consumption, carbon emissions, and resource utilization in data centers. By synthesizing insights from existing literature, this work explores strategies such as resource-efficient training, sustainable deployment practices, and lifecycle assessments to mitigate the environmental impacts of LLMs. Key areas of emphasis include energy optimization, renewable energy integration, and balancing performance with sustainability. The findings aim to guide researchers, practitioners, and policymakers in developing actionable strategies for sustainable AI systems, fostering a responsible and environmentally conscious future for artificial intelligence.
The rapid development of Artificial Intelligence (AI) has revolutionized numerous fields, with large language models (LLMs) and computer vision (CV) systems driving advancements in natural language understanding and visual processing, respectively. The convergence of these technologies has catalyzed the rise of multimodal AI, enabling richer, cross-modal understanding that spans text, vision, audio, and video modalities. Multimodal large language models (MLLMs), in particular, have emerged as a powerful framework, demonstrating impressive capabilities in tasks like image-text generation, visual question answering, and cross-modal retrieval. Despite these advancements, the complexity and scale of MLLMs introduce significant challenges in interpretability and explainability, essential for establishing transparency, trustworthiness, and reliability in high-stakes applications. This paper provides a comprehensive survey on the interpretability and explainability of MLLMs, proposing a novel framework that categorizes existing research across three perspectives: (I) Data, (II) Model, (III) Training \& Inference. We systematically analyze interpretability from token-level to embedding-level representations, assess approaches related to both architecture analysis and design, and explore training and inference strategies that enhance transparency. By comparing various methodologies, we identify their strengths and limitations and propose future research directions to address unresolved challenges in multimodal explainability. This survey offers a foundational resource for advancing interpretability and transparency in MLLMs, guiding researchers and practitioners toward developing more accountable and robust multimodal AI systems.
Vision-Language models like CLIP have been shown to be highly effective at linking visual perception and natural language understanding, enabling sophisticated image-text capabilities, including strong retrieval and zero-shot classification performance. Their widespread use, as well as the fact that CLIP models are trained on image-text pairs from the web, make them both a worthwhile and relatively easy target for backdoor attacks. As training foundational models, such as CLIP, from scratch is very expensive, this paper focuses on cleaning potentially poisoned models via fine-tuning. We first show that existing cleaning techniques are not effective against simple structured triggers used in Blended or BadNet backdoor attacks, exposing a critical vulnerability for potential real-world deployment of these models. Then, we introduce PAR, Perturb and Recover, a surprisingly simple yet effective mechanism to remove backdoors from CLIP models. Through extensive experiments across different encoders and types of backdoor attacks, we show that PAR achieves high backdoor removal rate while preserving good standard performance. Finally, we illustrate that our approach is effective even only with synthetic text-image pairs, i.e. without access to real training data. The code and models are available at https://github.com/nmndeep/PerturbAndRecover.
The growing capabilities of large language models in natural language understanding significantly strengthen existing agentic systems. To power performant on-device mobile agents for better data privacy, we introduce DroidCall, the first training and testing dataset for accurate Android intent invocation. With a highly flexible and reusable data generation pipeline, we constructed 10k samples in DroidCall. Given a task instruction in natural language, small language models such as Qwen2.5-3B and Gemma2-2B fine-tuned with DroidCall can approach or even surpass the capabilities of GPT-4o for accurate Android intent invocation. We also provide an end-to-end Android app equipped with these fine-tuned models to demonstrate the Android intent invocation process. The code and dataset are available at https://github.com/UbiquitousLearning/DroidCall.
The Nepali language has distinct linguistic features, especially its complex script (Devanagari script), morphology, and various dialects, which pose a unique challenge for natural language processing (NLP) evaluation. While the Nepali Language Understanding Evaluation (Nep-gLUE) benchmark provides a foundation for evaluating models, it remains limited in scope, covering four tasks. This restricts their utility for comprehensive assessments of NLP models. To address this limitation, we introduce eight new datasets, creating a new benchmark, the Nepali Language Understanding Evaluation (NLUE) benchmark, which covers a total of 12 tasks for evaluating the performance of models across a diverse set of Natural Language Understanding (NLU) tasks. The added tasks include single-sentence classification, similarity and paraphrase tasks, and Natural Language Inference (NLI) tasks. On evaluating the models using added tasks, we observe that the existing models fall short in handling complex NLU tasks effectively. This expanded benchmark sets a new standard for evaluating, comparing, and advancing models, contributing significantly to the broader goal of advancing NLP research for low-resource languages.
Zero-shot slot filling is a well-established subtask of Natural Language Understanding (NLU). However, most existing methods primarily focus on single-turn text data, overlooking the unique complexities of conversational dialogue. Conversational data is highly dynamic, often involving abrupt topic shifts, interruptions, and implicit references that make it difficult to directly apply zero-shot slot filling techniques, even with the remarkable capabilities of large language models (LLMs). This paper addresses these challenges by proposing strategies for automatic data annotation with slot induction and black-box knowledge distillation (KD) from a teacher LLM to a smaller model, outperforming vanilla LLMs on internal datasets by 26% absolute increase in F1 score. Additionally, we introduce an efficient system architecture for call center product settings that surpasses off-the-shelf extractive models by 34% relative F1 score, enabling near real-time inference on dialogue streams with higher accuracy, while preserving low latency.
GUIs have long been central to human-computer interaction, providing an intuitive and visually-driven way to access and interact with digital systems. The advent of LLMs, particularly multimodal models, has ushered in a new era of GUI automation. They have demonstrated exceptional capabilities in natural language understanding, code generation, and visual processing. This has paved the way for a new generation of LLM-brained GUI agents capable of interpreting complex GUI elements and autonomously executing actions based on natural language instructions. These agents represent a paradigm shift, enabling users to perform intricate, multi-step tasks through simple conversational commands. Their applications span across web navigation, mobile app interactions, and desktop automation, offering a transformative user experience that revolutionizes how individuals interact with software. This emerging field is rapidly advancing, with significant progress in both research and industry. To provide a structured understanding of this trend, this paper presents a comprehensive survey of LLM-brained GUI agents, exploring their historical evolution, core components, and advanced techniques. We address research questions such as existing GUI agent frameworks, the collection and utilization of data for training specialized GUI agents, the development of large action models tailored for GUI tasks, and the evaluation metrics and benchmarks necessary to assess their effectiveness. Additionally, we examine emerging applications powered by these agents. Through a detailed analysis, this survey identifies key research gaps and outlines a roadmap for future advancements in the field. By consolidating foundational knowledge and state-of-the-art developments, this work aims to guide both researchers and practitioners in overcoming challenges and unlocking the full potential of LLM-brained GUI agents.
Knowledge distillation (KD) has become a widely adopted approach for compressing large language models (LLMs) to reduce computational costs and memory footprints. However, the availability of complex teacher models is a prerequisite for running most KD pipelines. Thus, the traditional KD procedure can be unachievable or budget-unfriendly, particularly when relying on commercial LLMs like GPT4. In this regard, Self-distillation (SelfD) emerges as an advisable alternative, enabling student models to learn without teachers' guidance. Nonetheless, existing SelfD approaches for LMs often involve architectural modifications, assuming the models are open-source, which may not always be practical. In this work, we introduce a model-agnostic and task-agnostic method named dynamic SelfD from the previous minibatch (DynSDPB), which realizes current iterations' distillation from the last ones' generated logits. Additionally, to address prediction inaccuracies during the early iterations, we dynamically adjust the distillation influence and temperature values to enhance the adaptability of fine-tuning. Furthermore, DynSDPB is a novel fine-tuning policy that facilitates the seamless integration of existing self-correction and self-training techniques for small language models (SLMs) because they all require updating SLMs' parameters. We demonstrate the superior performance of DynSDPB on both encoder-only LMs (e.g., BERT model families) and decoder-only LMs (e.g., LLaMA model families), validating its effectiveness across natural language understanding (NLU) and natural language generation (NLG) benchmarks.
Referring Video Object Segmentation (RVOS) relies on natural language expressions to segment an object in a video clip. Existing methods restrict reasoning either to independent short clips, losing global context, or process the entire video offline, impairing their application in a streaming fashion. In this work, we aim to surpass these limitations and design an RVOS method capable of effectively operating in streaming-like scenarios while retaining contextual information from past frames. We build upon the Segment-Anything 2 (SAM2) model, that provides robust segmentation and tracking capabilities and is naturally suited for streaming processing. We make SAM2 wiser, by empowering it with natural language understanding and explicit temporal modeling at the feature extraction stage, without fine-tuning its weights, and without outsourcing modality interaction to external models. To this end, we introduce a novel adapter module that injects temporal information and multi-modal cues in the feature extraction process. We further reveal the phenomenon of tracking bias in SAM2 and propose a learnable module to adjust its tracking focus when the current frame features suggest a new object more aligned with the caption. Our proposed method, SAMWISE, achieves state-of-the-art across various benchmarks, by adding a negligible overhead of less than 5 M parameters. Code is available at https://github.com/ClaudiaCuttano/SAMWISE .
Visual Question Answering (VQA) is a challenge task that combines natural language processing and computer vision techniques and gradually becomes a benchmark test task in multimodal large language models (MLLMs). The goal of our survey is to provide an overview of the development of VQA and a detailed description of the latest models with high timeliness. This survey gives an up-to-date synthesis of natural language understanding of images and text, as well as the knowledge reasoning module based on image-question information on the core VQA tasks. In addition, we elaborate on recent advances in extracting and fusing modal information with vision-language pretraining models and multimodal large language models in VQA. We also exhaustively review the progress of knowledge reasoning in VQA by detailing the extraction of internal knowledge and the introduction of external knowledge. Finally, we present the datasets of VQA and different evaluation metrics and discuss possible directions for future work.
Pre-trained vision-language models (VLMs) have showcased remarkable performance in image and natural language understanding, such as image captioning and response generation. As the practical applications of vision-language models become increasingly widespread, their potential safety and robustness issues raise concerns that adversaries may evade the system and cause these models to generate toxic content through malicious attacks. Therefore, evaluating the robustness of open-source VLMs against adversarial attacks has garnered growing attention, with transfer-based attacks as a representative black-box attacking strategy. However, most existing transfer-based attacks neglect the importance of the semantic correlations between vision and text modalities, leading to sub-optimal adversarial example generation and attack performance. To address this issue, we present Chain of Attack (CoA), which iteratively enhances the generation of adversarial examples based on the multi-modal semantic update using a series of intermediate attacking steps, achieving superior adversarial transferability and efficiency. A unified attack success rate computing method is further proposed for automatic evasion evaluation. Extensive experiments conducted under the most realistic and high-stakes scenario, demonstrate that our attacking strategy can effectively mislead models to generate targeted responses using only black-box attacks without any knowledge of the victim models. The comprehensive robustness evaluation in our paper provides insight into the vulnerabilities of VLMs and offers a reference for the safety considerations of future model developments.
Real-time conversational AI agents face challenges in performing Natural Language Understanding (NLU) in dynamic, outdoor environments like automated drive-thru systems. These settings require NLU models to handle background noise, diverse accents, and multi-intent queries while operating under strict latency and memory constraints on edge devices. Additionally, robustness to errors from upstream Automatic Speech Recognition (ASR) is crucial, as ASR outputs in these environments are often noisy. We introduce Babylon, a transformer-based architecture that tackles NLU as an intent translation task, converting natural language inputs into sequences of regular language units ('transcodes') that encode both intents and slot information. This formulation allows Babylon to manage multi-intent scenarios in a single dialogue turn. Furthermore, Babylon incorporates an LSTM-based token pooling mechanism to preprocess phoneme sequences, reducing input length and optimizing for low-latency, low-memory edge deployment. This also helps mitigate inaccuracies in ASR outputs, enhancing system robustness. While this work focuses on drive-thru ordering, Babylon's design extends to similar noise-prone scenarios, for e.g. ticketing kiosks. Our experiments show that Babylon achieves significantly better accuracy-latency-memory footprint trade-offs over typically employed NMT models like Flan-T5 and BART, demonstrating its effectiveness for real-time NLU in edge deployment settings.
In the recent past, a popular way of evaluating natural language understanding (NLU), was to consider a model's ability to perform natural language inference (NLI) tasks. In this paper, we investigate if NLI tasks, that are rarely used for LLM evaluation, can still be informative for evaluating LLMs. Focusing on five different NLI benchmarks across six models of different scales, we investigate if they are able to discriminate models of different size and quality and how their accuracies develop during training. Furthermore, we investigate the extent to which the softmax distributions of models align with human distributions in cases where statements are ambiguous or vague. Overall, our results paint a positive picture for the NLI tasks: we find that they are able to discriminate well between models at various stages of training, yet are not (all) saturated. Furthermore, we find that while the similarity of model distributions with human label distributions increases with scale, it is still much higher than the similarity between two populations of humans, making it a potentially interesting statistic to consider.
The development of artificial intelligence systems capable of understanding and reasoning about complex real-world scenarios is a significant challenge. In this work we present a novel approach to enhance and exploit LLM reactive capability to address complex problems and interpret deeply contextual real-world meaning. We introduce a method and a tool for creating a multimodal, knowledge-augmented formal representation of meaning that combines the strengths of large language models with structured semantic representations. Our method begins with an image input, utilizing state-of-the-art large language models to generate a natural language description. This description is then transformed into an Abstract Meaning Representation (AMR) graph, which is formalized and enriched with logical design patterns, and layered semantics derived from linguistic and factual knowledge bases. The resulting graph is then fed back into the LLM to be extended with implicit knowledge activated by complex heuristic learning, including semantic implicatures, moral values, embodied cognition, and metaphorical representations. By bridging the gap between unstructured language models and formal semantic structures, our method opens new avenues for tackling intricate problems in natural language understanding and reasoning.
Personalized driving refers to an autonomous vehicle's ability to adapt its driving behavior or control strategies to match individual users' preferences and driving styles while maintaining safety and comfort standards. However, existing works either fail to capture every individual preference precisely or become computationally inefficient as the user base expands. Vision-Language Models (VLMs) offer promising solutions to this front through their natural language understanding and scene reasoning capabilities. In this work, we propose a lightweight yet effective on-board VLM framework that provides low-latency personalized driving performance while maintaining strong reasoning capabilities. Our solution incorporates a Retrieval-Augmented Generation (RAG)-based memory module that enables continuous learning of individual driving preferences through human feedback. Through comprehensive real-world vehicle deployment and experiments, our system has demonstrated the ability to provide safe, comfortable, and personalized driving experiences across various scenarios and significantly reduce takeover rates by up to 76.9%. To the best of our knowledge, this work represents the first end-to-end VLM-based motion control system in real-world autonomous vehicles.
The widely used ReLU is favored for its hardware efficiency, {as the implementation at inference is a one bit sign case,} yet suffers from issues such as the ``dying ReLU'' problem, where during training, neurons fail to activate and constantly remain at zero, as highlighted by Lu et al. Traditional approaches to mitigate this issue often introduce more complex and less hardware-friendly activation functions. In this work, we propose a Hysteresis Rectified Linear Unit (HeLU), an efficient activation function designed to address the ``dying ReLU'' problem with minimal complexity. Unlike traditional activation functions with fixed thresholds for training and inference, HeLU employs a variable threshold that refines the backpropagation. This refined mechanism allows simpler activation functions to achieve competitive performance comparable to their more complex counterparts without introducing unnecessary complexity or requiring inductive biases. Empirical evaluations demonstrate that HeLU enhances model generalization across diverse datasets, offering a promising solution for efficient and effective inference suitable for a wide range of neural network architectures.