Search

In recent times training Language Models (LMs) have relied on computationally heavy training over massive datasets which makes this training process extremely laborious. In this paper we propose a novel method for numerically evaluating text quality in large unlabelled NLP datasets in a model agnostic manner to assign the text instances a "quality score". By proposing the text quality metric, the paper establishes a framework to identify and eliminate low-quality text instances, leading to improved training efficiency for LM models. Experimental results over multiple models and datasets demonstrate the efficacy of this approach, showcasing substantial gains in training effectiveness and highlighting the potential for resource-efficient LM training. For example, we observe an absolute accuracy improvement of 0.9% averaged over 14 downstream evaluation tasks for multiple LM models while using 40% lesser data and training 42% faster when training on the OpenWebText dataset and 0.8% average absolute accuracy improvement while using 20% lesser data and training 21% faster on the Wikipedia dataset.

Model Weight Averaging (MWA) is a technique that seeks to enhance model's performance by averaging the weights of multiple trained models. This paper first empirically finds that 1) the vanilla MWA can benefit the class-imbalanced learning, and 2) performing model averaging in the early epochs of training yields a greater performance improvement than doing that in later epochs. Inspired by these two observations, in this paper we propose a novel MWA technique for class-imbalanced learning tasks named Iterative Model Weight Averaging (IMWA). Specifically, IMWA divides the entire training stage into multiple episodes. Within each episode, multiple models are concurrently trained from the same initialized model weight, and subsequently averaged into a singular model. Then, the weight of this average model serves as a fresh initialization for the ensuing episode, thus establishing an iterative learning paradigm. Compared to vanilla MWA, IMWA achieves higher performance improvements with the same computational cost. Moreover, IMWA can further enhance the performance of those methods employing EMA strategy, demonstrating that IMWA and EMA can complement each other. Extensive experiments on various class-imbalanced learning tasks, i.e., class-imbalanced image classification, semi-supervised class-imbalanced image classification and semi-supervised object detection tasks showcase the effectiveness of our IMWA.

When faced with accomplishing a task, human experts exhibit intentional behavior. Their unique intents shape their plans and decisions, resulting in experts demonstrating diverse behaviors to accomplish the same task. Due to the uncertainties encountered in the real world and their bounded rationality, experts sometimes adjust their intents, which in turn influences their behaviors during task execution. This paper introduces IDIL, a novel imitation learning algorithm to mimic these diverse intent-driven behaviors of experts. Iteratively, our approach estimates expert intent from heterogeneous demonstrations and then uses it to learn an intent-aware model of their behavior. Unlike contemporary approaches, IDIL is capable of addressing sequential tasks with high-dimensional state representations, while sidestepping the complexities and drawbacks associated with adversarial training (a mainstay of related techniques). Our empirical results suggest that the models generated by IDIL either match or surpass those produced by recent imitation learning benchmarks in metrics of task performance. Moreover, as it creates a generative model, IDIL demonstrates superior performance in intent inference metrics, crucial for human-agent interactions, and aptly captures a broad spectrum of expert behaviors., Comment: Extended version of an identically-titled paper accepted at AAMAS 2024

Fuzzing, a widely-used technique for bug detection, has seen advancements through Large Language Models (LLMs). Despite their potential, LLMs face specific challenges in fuzzing. In this paper, we identified five major challenges of LLM-assisted fuzzing. To support our findings, we revisited the most recent papers from top-tier conferences, confirming that these challenges are widespread. As a remedy, we propose some actionable recommendations to help improve applying LLM in Fuzzing and conduct preliminary evaluations on DBMS fuzzing. The results demonstrate that our recommendations effectively address the identified challenges.

In this paper, we present OmniSearchSage, a versatile and scalable system for understanding search queries, pins, and products for Pinterest search. We jointly learn a unified query embedding coupled with pin and product embeddings, leading to an improvement of $>8\%$ relevance, $>7\%$ engagement, and $>5\%$ ads CTR in Pinterest's production search system. The main contributors to these gains are improved content understanding, better multi-task learning, and real-time serving. We enrich our entity representations using diverse text derived from image captions from a generative LLM, historical engagement, and user-curated boards. Our multitask learning setup produces a single search query embedding in the same space as pin and product embeddings and compatible with pre-existing pin and product embeddings. We show the value of each feature through ablation studies, and show the effectiveness of a unified model compared to standalone counterparts. Finally, we share how these embeddings have been deployed across the Pinterest search stack, from retrieval to ranking, scaling to serve $300k$ requests per second at low latency. Our implementation of this work is available at https://github.com/pinterest/atg-research/tree/main/omnisearchsage., Comment: 8 pages, 5 figures, to be published as an oral paper in TheWebConf Industry Track 2024

As technology advances, the use of Machine Learning (ML) in cybersecurity is becoming increasingly crucial to tackle the growing complexity of cyber threats. While traditional ML models can enhance cybersecurity, their high energy and resource demands limit their applications, leading to the emergence of Tiny Machine Learning (TinyML) as a more suitable solution for resource-constrained environments. TinyML is widely applied in areas such as smart homes, healthcare, and industrial automation. TinyML focuses on optimizing ML algorithms for small, low-power devices, enabling intelligent data processing directly on edge devices. This paper provides a comprehensive review of common challenges of TinyML techniques, such as power consumption, limited memory, and computational constraints; it also explores potential solutions to these challenges, such as energy harvesting, computational optimization techniques, and transfer learning for privacy preservation. On the other hand, this paper discusses TinyML's applications in advancing cybersecurity for Electric Vehicle Charging Infrastructures (EVCIs) as a representative use case. It presents an experimental case study that enhances cybersecurity in EVCI using TinyML, evaluated against traditional ML in terms of reduced delay and memory usage, with a slight trade-off in accuracy. Additionally, the study includes a practical setup using the ESP32 microcontroller in the PlatformIO environment, which provides a hands-on assessment of TinyML's application in cybersecurity for EVCI., Comment: Submitted to IEEE Access; Code is available at GitHub link: https://github.com/Western-OC2-Lab/TinyML_EVCI

Large language model (LLM) services have recently begun offering a plugin ecosystem to interact with third-party API services. This innovation enhances the capabilities of LLMs, but it also introduces risks, as these plugins developed by various third parties cannot be easily trusted. This paper proposes a new attacking framework to examine security and safety vulnerabilities within LLM platforms that incorporate third-party services. Applying our framework specifically to widely used LLMs, we identify real-world malicious attacks across various domains on third-party APIs that can imperceptibly modify LLM outputs. The paper discusses the unique challenges posed by third-party API integration and offers strategic possibilities to improve the security and safety of LLM ecosystems moving forward. Our code is released at https://github.com/vk0812/Third-Party-Attacks-on-LLMs., Comment: ICLR 2024 Workshop on Secure and Trustworthy Large Language Models

In this paper, we propose a quantum computing oriented benchmark for combinatorial optimization. This benchmark, coined as QOPTLib, is composed of 40 instances equally distributed over four well-known problems: Traveling Salesman Problem, Vehicle Routing Problem, one-dimensional Bin Packing Problem and the Maximum Cut Problem. The sizes of the instances in QOPTLib not only correspond to computationally addressable sizes, but also to the maximum length approachable with non-zero likelihood of getting a good result. In this regard, it is important to highlight that hybrid approaches are also taken into consideration. Thus, this benchmark constitutes the first effort to provide users a general-purpose dataset. Also in this paper, we introduce a first full solving of QOPTLib using two solvers based on quantum annealing. Our main intention with this is to establish a preliminary baseline, hoping to inspire other researchers to beat these outcomes with newly proposed quantum-based algorithms., Comment: 15 pages, 3 figures, 1 table

This paper evaluated the effectiveness of using generative AI to simplify science communication and enhance public trust in science. By comparing lay summaries of journal articles from PNAS, yoked to those generated by AI, this work assessed linguistic simplicity across such summaries and public perceptions. Study 1a analyzed simplicity features of PNAS abstracts (scientific summaries) and significance statements (lay summaries), observing that lay summaries were indeed linguistically simpler, but effect size differences were small. Study 1b used GPT-4 to create significance statements based on paper abstracts and this more than doubled the average effect size without fine-tuning. Finally, Study 2 experimentally demonstrated that simply-written GPT summaries facilitated more favorable public perceptions of scientists (their credibility, trustworthiness) than more complexly-written human PNAS summaries. AI has the potential to engage scientific communities and the public via a simple language heuristic, advocating for its integration into scientific dissemination for a more informed society.

This paper presents Killkan, the first dataset for automatic speech recognition (ASR) in the Kichwa language, an indigenous language of Ecuador. Kichwa is an extremely low-resource endangered language, and there have been no resources before Killkan for Kichwa to be incorporated in applications of natural language processing. The dataset contains approximately 4 hours of audio with transcription, translation into Spanish, and morphosyntactic annotation in the format of Universal Dependencies. The audio data was retrieved from a publicly available radio program in Kichwa. This paper also provides corpus-linguistic analyses of the dataset with a special focus on the agglutinative morphology of Kichwa and frequent code-switching with Spanish. The experiments show that the dataset makes it possible to develop the first ASR system for Kichwa with reliable quality despite its small dataset size. This dataset, the ASR model, and the code used to develop them will be publicly available. Thus, our study positively showcases resource building and its applications for low-resource languages and their community., Comment: 11 pages, 9 tables, 3 figures, to be published in LREC-COLING 2024