Search

Understanding how language model performance varies with scale is critical to benchmark and algorithm development. Scaling laws are one approach to building this understanding, but the requirement of training models across many different scales has limited their use. We propose an alternative, observational approach that bypasses model training and instead builds scaling laws from ~80 publically available models. Building a single scaling law from multiple model families is challenging due to large variations in their training compute efficiencies and capabilities. However, we show that these variations are consistent with a simple, generalized scaling law where language model performance is a function of a low-dimensional capability space, and model families only vary in their efficiency in converting training compute to capabilities. Using this approach, we show the surprising predictability of complex scaling phenomena: we show that several emergent phenomena follow a smooth, sigmoidal behavior and are predictable from small models; we show that the agent performance of models such as GPT-4 can be precisely predicted from simpler non-agentic benchmarks; and we show how to predict the impact of post-training interventions like Chain-of-Thought and Self-Consistency as language model capabilities continue to improve.

The rapid development of Large Language Models (LLMs) demonstrates remarkable multilingual capabilities in natural language processing, attracting global attention in both academia and industry. To mitigate potential discrimination and enhance the overall usability and accessibility for diverse language user groups, it is important for the development of language-fair technology. Despite the breakthroughs of LLMs, the investigation into the multilingual scenario remains insufficient, where a comprehensive survey to summarize recent approaches, developments, limitations, and potential solutions is desirable. To this end, we provide a survey with multiple perspectives on the utilization of LLMs in the multilingual scenario. We first rethink the transitions between previous and current research on pre-trained language models. Then we introduce several perspectives on the multilingualism of LLMs, including training and inference methods, model security, multi-domain with language culture, and usage of datasets. We also discuss the major challenges that arise in these aspects, along with possible solutions. Besides, we highlight future research directions that aim at further enhancing LLMs with multilingualism. The survey aims to help the research community address multilingual problems and provide a comprehensive understanding of the core concepts, key techniques, and latest developments in multilingual natural language processing based on LLMs., Comment: 54 pages, Work in Progress

Self-distillation methods using Siamese networks are popular for self-supervised pre-training. DINO is one such method based on a cross-entropy loss between $K$-dimensional probability vectors, obtained by applying a softmax function to the dot product between representations and learnt prototypes. Given the fact that the learned representations are $L^2$-normalized, we show that DINO and its derivatives, such as iBOT, can be interpreted as a mixture model of von Mises-Fisher components. With this interpretation, DINO assumes equal precision for all components when the prototypes are also $L^2$-normalized. Using this insight we propose DINO-vMF, that adds appropriate normalization constants when computing the cluster assignment probabilities. Unlike DINO, DINO-vMF is stable also for the larger ViT-Base model with unnormalized prototypes. We show that the added flexibility of the mixture model is beneficial in terms of better image representations. The DINO-vMF pre-trained model consistently performs better than DINO on a range of downstream tasks. We obtain similar improvements for iBOT-vMF vs iBOT and thereby show the relevance of our proposed modification also for other methods derived from DINO., Comment: Accepted to ICLR 2023

Multiple imputation (MI) models can be improved by including auxiliary covariates (AC), but their performance in high-dimensional data is not well understood. We aimed to develop and compare high-dimensional MI (HDMI) approaches using structured and natural language processing (NLP)-derived AC in studies with partially observed confounders. We conducted a plasmode simulation study using data from opioid vs. non-steroidal anti-inflammatory drug (NSAID) initiators (X) with observed serum creatinine labs (Z2) and time-to-acute kidney injury as outcome. We simulated 100 cohorts with a null treatment effect, including X, Z2, atrial fibrillation (U), and 13 other investigator-derived confounders (Z1) in the outcome generation. We then imposed missingness (MZ2) on 50% of Z2 measurements as a function of Z2 and U and created different HDMI candidate AC using structured and NLP-derived features. We mimicked scenarios where U was unobserved by omitting it from all AC candidate sets. Using LASSO, we data-adaptively selected HDMI covariates associated with Z2 and MZ2 for MI, and with U to include in propensity score models. The treatment effect was estimated following propensity score matching in MI datasets and we benchmarked HDMI approaches against a baseline imputation and complete case analysis with Z1 only. HDMI using claims data showed the lowest bias (0.072). Combining claims and sentence embeddings led to an improvement in the efficiency displaying the lowest root-mean-squared-error (0.173) and coverage (94%). NLP-derived AC alone did not perform better than baseline MI. HDMI approaches may decrease bias in studies with partially observed confounders where missingness depends on unobserved factors.

In the e-commerce domain, the accurate extraction of attribute-value pairs from product listings (e.g., Brand: Apple) is crucial for enhancing search and recommendation systems. The automation of this extraction process is challenging due to the vast diversity of product categories and their respective attributes, compounded by the lack of extensive, accurately annotated training datasets and the demand for low latency to meet the real-time needs of e-commerce platforms. To address these challenges, we introduce GenToC, a novel two-stage model for extracting attribute-value pairs from product titles. GenToC is designed to train with partially-labeled data, leveraging incomplete attribute-value pairs and obviating the need for a fully annotated dataset. Moreover, we introduce a bootstrapping method that enables GenToC to progressively refine and expand its training dataset. This enhancement substantially improves the quality of data available for training other neural network models that are typically faster but are inherently less capable than GenToC in terms of their capacity to handle partially-labeled data. By supplying an enriched dataset for training, GenToC significantly advances the performance of these alternative models, making them more suitable for real-time deployment. Our results highlight the unique capability of GenToC to learn from a limited set of labeled data and to contribute to the training of more efficient models, marking a significant leap forward in the automated extraction of attribute-value pairs from product titles. GenToC has been successfully integrated into India's largest B2B e-commerce platform, IndiaMART.com, achieving a significant increase of 21.1% in recall over the existing deployed system while maintaining a high precision of 89.5% in this challenging task.

Large Language Models (LLMs) constitute a breakthrough state-of-the-art Artificial Intelligence (AI) technology which is rapidly evolving and promises to aid in medical diagnosis either by assisting doctors or by simulating a doctor's workflow in more advanced and complex implementations. In this technical paper, we outline Cognitive Network Evaluation Toolkit for Medical Domains (COGNET-MD), which constitutes a novel benchmark for LLM evaluation in the medical domain. Specifically, we propose a scoring-framework with increased difficulty to assess the ability of LLMs in interpreting medical text. The proposed framework is accompanied with a database of Multiple Choice Quizzes (MCQs). To ensure alignment with current medical trends and enhance safety, usefulness, and applicability, these MCQs have been constructed in collaboration with several associated medical experts in various medical domains and are characterized by varying degrees of difficulty. The current (first) version of the database includes the medical domains of Psychiatry, Dentistry, Pulmonology, Dermatology and Endocrinology, but it will be continuously extended and expanded to include additional medical domains., Comment: Technical Paper

The exponential growth of astronomical datasets provides an unprecedented opportunity for humans to gain insight into the Universe. However, effectively analyzing this vast amount of data poses a significant challenge. Astronomers are turning to deep learning techniques to address this, but the methods are limited by their specific training sets, leading to considerable duplicate workloads too. Hence, as an example to present how to overcome the issue, we built a framework for general analysis of galaxy images, based on a large vision model (LVM) plus downstream tasks (DST), including galaxy morphological classification, image restoration, object detection, parameter extraction, and more. Considering the low signal-to-noise ratio of galaxy images and the imbalanced distribution of galaxy categories, we have incorporated a Human-in-the-loop (HITL) module into our large vision model, which leverages human knowledge to enhance the reliability and interpretability of processing galaxy images interactively. The proposed framework exhibits notable few-shot learning capabilities and versatile adaptability to all the abovementioned tasks on galaxy images in the DESI legacy imaging surveys. Expressly, for object detection, trained by 1000 data points, our DST upon the LVM achieves an accuracy of 96.7%, while ResNet50 plus Mask R-CNN gives an accuracy of 93.1%; for morphology classification, to obtain AUC ~0.9, LVM plus DST and HITL only requests 1/50 training sets compared to ResNet18. Expectedly, multimodal data can be integrated similarly, which opens up possibilities for conducting joint analyses with datasets spanning diverse domains in the era of multi-message astronomy., Comment: 26 pages, 10 figures, to be published on Chinese Physics C

This review aims to systematically assess the current status and prospects of artificial intelligence (AI) in the rehabilitation management of patients with schizophrenia and their impact on the rehabilitation process. We selected 70 studies from 2012 to the present, focusing on application, technology categories, products, and data types of machine learning, deep learning, reinforcement learning, and other technologies in mental health interventions and management. The results indicate that AI can be widely used in symptom monitoring, relapse risk prediction, and rehabilitation treatment by analyzing ecological momentary assessment, behavioral, and speech data. This review further explores the potential challenges and future directions of emerging products, technologies, and analytical methods based on AI, such as social media analysis, serious games, and large language models in rehabilitation. In summary, this study systematically reviews the application status of AI in schizophrenia rehabilitation management and provides valuable insights and recommendations for future research paths.

Time series (TS) forecasting has been an unprecedentedly popular problem in recent years, with ubiquitous applications in both scientific and business fields. Various approaches have been introduced to time series analysis, including both statistical approaches and deep neural networks. Although neural network approaches have illustrated stronger ability of representation than statistical methods, they struggle to provide sufficient interpretablility, and can be too complicated to optimize. In this paper, we present WEITS, a frequency-aware deep learning framework that is highly interpretable and computationally efficient. Through multi-level wavelet decomposition, WEITS novelly infuses frequency analysis into a highly deep learning framework. Combined with a forward-backward residual architecture, it enjoys both high representation capability and statistical interpretability. Extensive experiments on real-world datasets have demonstrated competitive performance of our model, along with its additional advantage of high computation efficiency. Furthermore, WEITS provides a general framework that can always seamlessly integrate with state-of-the-art approaches for time series forecast., Comment: arXiv admin note: text overlap with arXiv:2310.09488 by other authors

One way to personalize chatbot interactions is by establishing common ground with the intended reader. A domain where establishing mutual understanding could be particularly impactful is vaccine concerns and misinformation. Vaccine interventions are forms of messaging which aim to answer concerns expressed about vaccination. Tailoring responses in this domain is difficult, since opinions often have seemingly little ideological overlap. We define the task of tailoring vaccine interventions to a Common-Ground Opinion (CGO). Tailoring responses to a CGO involves meaningfully improving the answer by relating it to an opinion or belief the reader holds. In this paper we introduce TAILOR-CGO, a dataset for evaluating how well responses are tailored to provided CGOs. We benchmark several major LLMs on this task; finding GPT-4-Turbo performs significantly better than others. We also build automatic evaluation metrics, including an efficient and accurate BERT model that outperforms finetuned LLMs, investigate how to successfully tailor vaccine messaging to CGOs, and provide actionable recommendations from this investigation. Code and model weights: https://github.com/rickardstureborg/tailor-cgo Dataset: https://huggingface.co/datasets/DukeNLP/tailor-cgo, Comment: NAACL Findings 2024

Generative pre-trained large language models (LLMs) have demonstrated impressive performance over a wide range of tasks, thanks to the unprecedented amount of data they have been trained on. As established scaling laws indicate, LLMs' future performance improvement depends on the amount of computing and data sources we can leverage for pre-training. Federated learning (FL) has the potential to unleash the majority of the planet's data and computational resources, which are underutilized by the data-center-focused training methodology of current LLM practice. Our work presents a robust, flexible, reproducible FL approach that enables large-scale collaboration across institutions to train LLMs. This would mobilize more computational and data resources while matching or potentially exceeding centralized performance. We further show the effectiveness of the federated training scales with model size and present our approach for training a billion-scale federated LLM using limited resources. This will help data-rich actors to become the protagonists of LLMs pre-training instead of leaving the stage to compute-rich actors alone., Comment: 10 pages, 4 figures, pre-print

The Shapley value (SV) is a prevalent approach of allocating credit to machine learning (ML) entities to understand black box ML models. Enriching such interpretations with higher-order interactions is inevitable for complex systems, where the Shapley Interaction Index (SII) is a direct axiomatic extension of the SV. While it is well-known that the SV yields an optimal approximation of any game via a weighted least square (WLS) objective, an extension of this result to SII has been a long-standing open problem, which even led to the proposal of an alternative index. In this work, we characterize higher-order SII as a solution to a WLS problem, which constructs an optimal approximation via SII and $k$-Shapley values ($k$-SII). We prove this representation for the SV and pairwise SII and give empirically validated conjectures for higher orders. As a result, we propose KernelSHAP-IQ, a direct extension of KernelSHAP for SII, and demonstrate state-of-the-art performance for feature interactions., Comment: Accepted Paper at ICML 2024. This version is not the Camera Ready Version

Deep learning models have performed well on many NLP tasks. However, their internal mechanisms are typically difficult for humans to understand. The development of methods to explain models has become a key issue in the reliability of deep learning models in many important applications. Various saliency explanation methods, which give each feature of input a score proportional to the contribution of output, have been proposed to determine the part of the input which a model values most. Despite a considerable body of work on the evaluation of saliency methods, whether the results of various evaluation metrics agree with human cognition remains an open question. In this study, we propose a new human-based method to evaluate saliency methods in NLP by crowdsourcing. We recruited 800 crowd workers and empirically evaluated seven saliency methods on two datasets with the proposed method. We analyzed the performance of saliency methods, compared our results with existing automated evaluation methods, and identified notable differences between NLP and computer vision (CV) fields when using saliency methods. The instance-level data of our crowdsourced experiments and the code to reproduce the explanations are available at https://github.com/xtlu/lreccoling_evaluation., Comment: 13 pages, 4 figures, Accepted for LREC-Coling 2024 (Oral)

Partially observable Markov Decision Processes (POMDPs) are a standard model for agents making decisions in uncertain environments. Most work on POMDPs focuses on synthesizing strategies based on the available capabilities. However, system designers can often control an agent's observation capabilities, e.g. by placing or selecting sensors. This raises the question of how one should select an agent's sensors cost-effectively such that it achieves the desired goals. In this paper, we study the novel optimal observability problem OOP: Given a POMDP M, how should one change M's observation capabilities within a fixed budget such that its (minimal) expected reward remains below a given threshold? We show that the problem is undecidable in general and decidable when considering positional strategies only. We present two algorithms for a decidable fragment of the OOP: one based on optimal strategies of M's underlying Markov decision process and one based on parameter synthesis with SMT. We report promising results for variants of typical examples from the POMDP literature.

In the realm of globalized financial markets, commercial banks are confronted with an escalating magnitude of credit risk, thereby imposing heightened requisites upon the security of bank assets and financial stability. This study harnesses advanced neural network techniques, notably the Backpropagation (BP) neural network, to pioneer a novel model for preempting credit risk in commercial banks. The discourse initially scrutinizes conventional financial risk preemptive models, such as ARMA, ARCH, and Logistic regression models, critically analyzing their real-world applications. Subsequently, the exposition elaborates on the construction process of the BP neural network model, encompassing network architecture design, activation function selection, parameter initialization, and objective function construction. Through comparative analysis, the superiority of neural network models in preempting credit risk in commercial banks is elucidated. The experimental segment selects specific bank data, validating the model's predictive accuracy and practicality. Research findings evince that this model efficaciously enhances the foresight and precision of credit risk management.

Artificially sweetened beverages like Diet Coke are often considered healthier alternatives, but the debate over their impact on obesity persists. Previous research has predominantly relied on observational data or randomized controlled trials (RCTs), which may not accurately capture the causal relationship between Diet Coke consumption and obesity. This study uses causal inference methods, employing data from the National Health and Nutrition Examination Survey (NHANES) to examine this relationship across diverse demographics. Instead of relying on RCT data, we constructed a causal graph and applied the back-door criterion with its adjustment formula to estimate the RCT distributions. We then calculated the counterfactual quantity, the Probability of Necessity and Sufficiency (PNS), using both NHANES data and estimated RCT data. We propose that PNS is the essential metric for assessing the impact of Diet Coke on obesity. Our results indicate that between 20% to 50% of individuals, especially those with poor dietary habits, are more likely to gain weight from Diet Coke. Conversely, in groups like young females with healthier diets, only a small proportion experience weight gain due to Diet Coke. These findings highlight the influence of individual lifestyle and potential hormonal factors on the varied effects of Diet Coke, providing a new framework for understanding its nutritional impacts on health., Comment: 9 pages, 4 figures, uses aaai24.sty

Knowledge-intensive tasks pose a significant challenge for Machine Learning (ML) techniques. Commonly adopted methods, such as Large Language Models (LLMs), often exhibit limitations when applied to such tasks. Nevertheless, there have been notable endeavours to mitigate these challenges, with a significant emphasis on augmenting LLMs through Knowledge Graphs (KGs). While KGs provide many advantages for representing knowledge, their development costs can deter extensive research and applications. Addressing this limitation, we introduce a framework for enriching embeddings of small-scale domain-specific Knowledge Graphs with well-established general-purpose KGs. Adopting our method, a modest domain-specific KG can benefit from a performance boost in downstream tasks when linked to a substantial general-purpose KG. Experimental evaluations demonstrate a notable enhancement, with up to a 44% increase observed in the Hits@10 metric. This relatively unexplored research direction can catalyze more frequent incorporation of KGs in knowledge-intensive tasks, resulting in more robust, reliable ML implementations, which hallucinates less than prevalent LLM solutions. Keywords: knowledge graph, knowledge graph completion, entity alignment, representation learning, machine learning, Comment: Accepted for LREC-COLING 2024

In the past year, Multimodal Large Language Models (MLLMs) have demonstrated remarkable performance in tasks such as visual question answering, visual understanding and reasoning. However, the extensive model size and high training and inference costs have hindered the widespread application of MLLMs in academia and industry. Thus, studying efficient and lightweight MLLMs has enormous potential, especially in edge computing scenarios. In this survey, we provide a comprehensive and systematic review of the current state of efficient MLLMs. Specifically, we summarize the timeline of representative efficient MLLMs, research state of efficient structures and strategies, and the applications. Finally, we discuss the limitations of current efficient MLLM research and promising future directions. Please refer to our GitHub repository for more details: https://github.com/lijiannuist/Efficient-Multimodal-LLMs-Survey.

AI has become pervasive in recent years, but state-of-the-art approaches predominantly neglect the need for AI systems to be contestable. Instead, contestability is advocated by AI guidelines (e.g. by the OECD) and regulation of automated decision-making (e.g. GDPR). In this position paper we explore how contestability can be achieved computationally in and for AI. We argue that contestable AI requires dynamic (human-machine and/or machine-machine) explainability and decision-making processes, whereby machines can (i) interact with humans and/or other machines to progressively explain their outputs and/or their reasoning as well as assess grounds for contestation provided by these humans and/or other machines, and (ii) revise their decision-making processes to redress any issues successfully raised during contestation. Given that much of the current AI landscape is tailored to static AIs, the need to accommodate contestability will require a radical rethinking, that, we argue, computational argumentation is ideally suited to support.

Coreference resolution, critical for identifying textual entities referencing the same entity, faces challenges in pronoun resolution, particularly identifying pronoun antecedents. Existing methods often treat pronoun resolution as a separate task from mention detection, potentially missing valuable information. This study proposes the first end-to-end neural network system for Persian pronoun resolution, leveraging pre-trained Transformer models like ParsBERT. Our system jointly optimizes both mention detection and antecedent linking, achieving a 3.37 F1 score improvement over the previous state-of-the-art system (which relied on rule-based and statistical methods) on the Mehr corpus. This significant improvement demonstrates the effectiveness of combining neural networks with linguistic models, potentially marking a significant advancement in Persian pronoun resolution and paving the way for further research in this under-explored area.

Drought is a complex environmental phenomenon that affects millions of people and communities all over the globe and is too elusive to be accurately predicted. This is mostly due to the scalability and variability of the web of environmental parameters that directly/indirectly causes the onset of different categories of drought. Since the dawn of man, efforts have been made to uniquely understand the natural indicators that provide signs of likely environmental events. These indicators/signs in the form of indigenous knowledge system have been used for generations. The intricate complexity of drought has, however, always been a major stumbling block for accurate drought prediction and forecasting systems. Recently, scientists in the field of agriculture and environmental monitoring have been discussing the integration of indigenous knowledge and scientific knowledge for a more accurate environmental forecasting system in order to incorporate diverse environmental information for a reliable drought forecast. Hence, in this research, the core objective is the development of a semantics-based data integration middleware that encompasses and integrates heterogeneous data models of local indigenous knowledge and sensor data towards an accurate drought forecasting system for the study areas. The local indigenous knowledge on drought gathered from the domain experts is transformed into rules to be used for performing deductive inference in conjunction with sensors data for determining the onset of drought through an automated inference generation module of the middleware. The semantic middleware incorporates, inter alia, a distributed architecture that consists of a streaming data processing engine based on Apache Kafka for real-time stream processing; a rule-based reasoning module; an ontology module for semantic representation of the knowledge bases., Comment: 286 Pages

Diaspora communities are disproportionately impacted by off-the-radar misinformation and often neglected by mainstream fact-checking efforts, creating a critical need to scale-up efforts of nascent fact-checking initiatives. In this paper we present SynDy, a framework for Synthetic Dynamic Dataset Generation to leverage the capabilities of the largest frontier Large Language Models (LLMs) to train local, specialized language models. To the best of our knowledge, SynDy is the first paper utilizing LLMs to create fine-grained synthetic labels for tasks of direct relevance to misinformation mitigation, namely Claim Matching, Topical Clustering, and Claim Relationship Classification. SynDy utilizes LLMs and social media queries to automatically generate distantly-supervised, topically-focused datasets with synthetic labels on these three tasks, providing essential tools to scale up human-led fact-checking at a fraction of the cost of human-annotated data. Training on SynDy's generated labels shows improvement over a standard baseline and is not significantly worse compared to training on human labels (which may be infeasible to acquire). SynDy is being integrated into Meedan's chatbot tiplines that are used by over 50 organizations, serve over 230K users annually, and automatically distribute human-written fact-checks via messaging apps such as WhatsApp. SynDy will also be integrated into our deployed Co-Insights toolkit, enabling low-resource organizations to launch tiplines for their communities. Finally, we envision SynDy enabling additional fact-checking tools such as matching new misinformation claims to high-quality explainers on common misinformation topics.

Our study focuses on comparing the performance and resource requirements between different Long Short-Term Memory (LSTM) neural network architectures and an ANN specialized architecture for forex market prediction. We analyze the execution time of the models as well as the resources consumed, such as memory and computational power. Our aim is to demonstrate that the specialized architecture not only achieves better results in forex market prediction but also executes using fewer resources and in a shorter time frame compared to LSTM architectures. This comparative analysis will provide significant insights into the suitability of these two types of architectures for time series prediction in the forex market environment.

With impressive achievements made, artificial intelligence is on the path forward to artificial general intelligence. Sora, developed by OpenAI, which is capable of minute-level world-simulative abilities can be considered as a milestone on this developmental path. However, despite its notable successes, Sora still encounters various obstacles that need to be resolved. In this survey, we embark from the perspective of disassembling Sora in text-to-video generation, and conducting a comprehensive review of literature, trying to answer the question, \textit{From Sora What We Can See}. Specifically, after basic preliminaries regarding the general algorithms are introduced, the literature is categorized from three mutually perpendicular dimensions: evolutionary generators, excellent pursuit, and realistic panorama. Subsequently, the widely used datasets and metrics are organized in detail. Last but more importantly, we identify several challenges and open problems in this domain and propose potential future directions for research and development., Comment: A comprehensive list of text-to-video generation studies in this survey is available at https://github.com/soraw-ai/Awesome-Text-to-Video-Generation

Large language models (LLMs) have become integral to our professional workflows and daily lives. Nevertheless, these machine companions of ours have a critical flaw: the huge amount of data which endows them with vast and diverse knowledge, also exposes them to the inevitable toxicity and bias. While most LLMs incorporate defense mechanisms to prevent the generation of harmful content, these safeguards can be easily bypassed with minimal prompt engineering. In this paper, we introduce the new Thoroughly Engineered Toxicity (TET) dataset, comprising manually crafted prompts designed to nullify the protective layers of such models. Through extensive evaluations, we demonstrate the pivotal role of TET in providing a rigorous benchmark for evaluation of toxicity awareness in several popular LLMs: it highlights the toxicity in the LLMs that might remain hidden when using normal prompts, thus revealing subtler issues in their behavior.

Federated Learning (FL) has gained attention for addressing data scarcity and privacy concerns. While parallel FL algorithms like FedAvg exhibit remarkable performance, they face challenges in scenarios with diverse network speeds and concerns about centralized control, especially in multi-institutional collaborations like the medical domain. Serial FL presents an alternative solution, circumventing these challenges by transferring model updates serially between devices in a cyclical manner. Nevertheless, it is deemed inferior to parallel FL in that (1) its performance shows undesirable fluctuations, and (2) it converges to a lower plateau, particularly when dealing with non-IID data. The observed phenomenon is attributed to catastrophic forgetting due to knowledge loss from previous sites. In this paper, to overcome fluctuation and low efficiency in the iterative learning and forgetting process, we introduce cyclical weight consolidation (CWC), a straightforward yet potent approach specifically tailored for serial FL. CWC employs a consolidation matrix to regulate local optimization. This matrix tracks the significance of each parameter on the overall federation throughout the entire training trajectory, preventing abrupt changes in significant weights. During revisitation, to maintain adaptability, old memory undergoes decay to incorporate new information. Our comprehensive evaluations demonstrate that in various non-IID settings, CWC mitigates the fluctuation behavior of the original serial FL approach and enhances the converged performance consistently and significantly. The improved performance is either comparable to or better than the parallel vanilla., Comment: 12 pages, 8 figures

In the era of exponential technology growth, one unexpected guest has claimed a seat in classrooms worldwide, Artificial Intelligence. Generative AI, such as ChatGPT, promises a revolution in education, yet it arrives with a double-edged sword. Its potential for personalized learning is offset by issues of cheating, inaccuracies, and educators struggling to incorporate it effectively into their lesson design. We are standing on the brink of this educational frontier, and it is clear that we need to navigate this terrain with a lot of care. This is a major challenge that could undermine the integrity and value of our educational process. So, how can we turn these challenges into opportunities? When used inappropriately, AI tools can become the perfect tool for the cut copy paste mentality, and quickly begin to corrode critical thinking, creativity, and deep understanding, the most important skills in our rapidly changing world. Teachers feel that they are not equipped to leverage this technology, widening the digital divide among educators and institutions. Addressing these concerns calls for an in depth research approach. We will employ empirical research, drawing on the Technology Acceptance Model, to assess the attitudes toward generative AI among educators and students. Understanding their perceptions, usage patterns, and hurdles is the first crucial step in creating an effective solution. The present study will be used as a process manual for future researchers to apply, running their own data, based on the steps explained here, Comment: 5 pages, 1 figure, 1 table

Model evaluations are central to understanding the safety, risks, and societal impacts of AI systems. While most real-world AI applications involve human-AI interaction, most current evaluations (e.g., common benchmarks) of AI models do not. Instead, they incorporate human factors in limited ways, assessing the safety of models in isolation, thereby falling short of capturing the complexity of human-model interactions. In this paper, we discuss and operationalize a definition of an emerging category of evaluations -- "human interaction evaluations" (HIEs) -- which focus on the assessment of human-model interactions or the process and the outcomes of humans using models. First, we argue that HIEs can be used to increase the validity of safety evaluations, assess direct human impact and interaction-specific harms, and guide future assessments of models' societal impact. Second, we propose a safety-focused HIE design framework -- containing a human-LLM interaction taxonomy -- with three stages: (1) identifying the risk or harm area, (2) characterizing the use context, and (3) choosing the evaluation parameters. Third, we apply our framework to two potential evaluations for overreliance and persuasion risks. Finally, we conclude with tangible recommendations for addressing concerns over costs, replicability, and unrepresentativeness of HIEs., Comment: 15 pages, 1 figure

The Multigenerator, Multidomain, and Multilingual Black-Box Machine-Generated Text Detection shared task in the SemEval-2024 competition aims to tackle the problem of misusing collaborative human-AI writing. Although there are a lot of existing detectors of AI content, they are often designed to give a binary answer and thus may not be suitable for more nuanced problem of finding the boundaries between human-written and machine-generated texts, while hybrid human-AI writing becomes more and more popular. In this paper, we address the boundary detection problem. Particularly, we present a pipeline for augmenting data for supervised fine-tuning of DeBERTaV3. We receive new best MAE score, according to the leaderboard of the competition, with this pipeline., Comment: New best score from the leaderboard, to appear in SemEval-2024 Workshop proceedings

This paper surveys and organizes research works on medical dialog systems, which is an important yet challenging task. Although these systems have been surveyed in the medical community from an application perspective, a systematic review from a rigorous technical perspective has to date remained noticeably absent. As a result, an overview of the categories, methods, and evaluation of medical dialogue systems remain limited and underspecified, hindering the further improvement of this area. To fill this gap, we investigate an initial pool of 325 papers from well-known computer science, and natural language processing conferences and journals, and make an overview. Recently, large language models have shown strong model capacity on downstream tasks, which also reshaped medical dialog systems' foundation. Despite the alluring practical application value, current medical dialogue systems still suffer from problems. To this end, this paper lists the grand challenges of medical dialog systems, especially of large language models.

We consider a constrained Markov Decision Problem (CMDP) where the goal of an agent is to maximize the expected discounted sum of rewards over an infinite horizon while ensuring that the expected discounted sum of costs exceeds a certain threshold. Building on the idea of momentum-based acceleration, we develop the Primal-Dual Accelerated Natural Policy Gradient (PD-ANPG) algorithm that guarantees an $\epsilon$ global optimality gap and $\epsilon$ constraint violation with $\mathcal{O}(\epsilon^{-3})$ sample complexity. This improves the state-of-the-art sample complexity in CMDP by a factor of $\mathcal{O}(\epsilon^{-1})$.

In the Vision-and-Language Navigation (VLN) task, the agent is required to navigate to a destination following a natural language instruction. While learning-based approaches have been a major solution to the task, they suffer from high training costs and lack of interpretability. Recently, Large Language Models (LLMs) have emerged as a promising tool for VLN due to their strong generalization capabilities. However, existing LLM-based methods face limitations in memory construction and diversity of navigation strategies. To address these challenges, we propose a suite of techniques. Firstly, we introduce a method to maintain a topological map that stores navigation history, retaining information about viewpoints, objects, and their spatial relationships. This map also serves as a global action space. Additionally, we present a Navigation Chain of Thoughts module, leveraging human navigation examples to enrich navigation strategy diversity. Finally, we establish a pipeline that integrates navigational memory and strategies with perception and action prediction modules. Experimental results on the REVERIE and R2R datasets show that our method effectively enhances the navigation ability of the LLM and improves the interpretability of navigation reasoning.

Transformer-based encoder-decoder models have demonstrated impressive results in chemical reaction prediction tasks. However, these models typically rely on pretraining using tens of millions of unlabelled molecules, which can be time-consuming and GPU-intensive. One of the central questions we aim to answer in this work is: Can FlanT5 and ByT5, the encode-decoder models pretrained solely on language data, be effectively specialised for organic reaction prediction through task-specific fine-tuning? We conduct a systematic empirical study on several key issues of the process, including tokenisation, the impact of (SMILES-oriented) pretraining, fine-tuning sample efficiency, and decoding algorithms at inference. Our key findings indicate that although being pretrained only on language tasks, FlanT5 and ByT5 provide a solid foundation to fine-tune for reaction prediction, and thus become `chemistry domain compatible' in the process. This suggests that GPU-intensive and expensive pretraining on a large dataset of unlabelled molecules may be useful yet not essential to leverage the power of language models for chemistry. All our models achieve comparable Top-1 and Top-5 accuracy although some variation across different models does exist. Notably, tokenisation and vocabulary trimming slightly affect final performance but can speed up training and inference; The most efficient greedy decoding strategy is very competitive while only marginal gains can be achieved from more sophisticated decoding algorithms. In summary, we evaluate FlanT5 and ByT5 across several dimensions and benchmark their impact on organic reaction prediction, which may guide more effective use of these state-of-the-art language models for chemistry-related tasks in the future., Comment: Preprint

Temporal Knowledge Graph (TKG) reasoning focuses on predicting events through historical information within snapshots distributed on a timeline. Existing studies mainly concentrate on two perspectives of leveraging the history of TKGs, including capturing evolution of each recent snapshot or correlations among global historical facts. Despite the achieved significant accomplishments, these models still fall short of (1) investigating the influences of multi-granularity interactions across recent snapshots and (2) harnessing the expressive semantics of significant links accorded with queries throughout the entire history, especially events exerting a profound impact on the future. These inadequacies restrict representation ability to reflect historical dependencies and future trends thoroughly. To overcome these drawbacks, we propose an innovative TKG reasoning approach towards \textbf{His}torically \textbf{R}elevant \textbf{E}vents \textbf{S}tructuring ($\mathsf{HisRES}$). Concretely, $\mathsf{HisRES}$ comprises two distinctive modules excelling in structuring historically relevant events within TKGs, including a multi-granularity evolutionary encoder that captures structural and temporal dependencies of the most recent snapshots, and a global relevance encoder that concentrates on crucial correlations among events relevant to queries from the entire history. Furthermore, $\mathsf{HisRES}$ incorporates a self-gating mechanism for adaptively merging multi-granularity recent and historically relevant structuring representations. Extensive experiments on four event-based benchmarks demonstrate the state-of-the-art performance of $\mathsf{HisRES}$ and indicate the superiority and effectiveness of structuring historical relevance for TKG reasoning.

Recent advances in the verification of deep neural networks (DNNs) have opened the way for broader usage of DNN verification technology in many application areas, including safety-critical ones. DNN verifiers are themselves complex programs that have been shown to be susceptible to errors and imprecisions; this in turn has raised the question of trust in DNN verifiers. One prominent attempt to address this issue is enhancing DNN verifiers with the capability of producing proofs of their results that are subject to independent algorithmic certification (proof checking). Formulations of proof production and proof checking already exist on top of the state-of-the-art Marabou DNN verifier. The native implementation of the proof checking algorithm for Marabou was done in C++ and itself raised the question of trust in the code (e.g., in the precision of floating point calculations or guarantees for implementation soundness). Here, we present an alternative implementation of the Marabou proof checking algorithm in Imandra -- an industrial functional programming language and prover -- that allows us to obtain an implementation with formal guarantees, including proofs of mathematical results underlying the algorithm, such as the use of the Farkas lemma.

Deep learning methods for time series have already reached excellent performances in both prediction and classification tasks, including anomaly detection. However, the complexity inherent in Cyber Physical Systems (CPS) creates a challenge when it comes to explainability methods. To overcome this inherent lack of interpretability, we propose ECATS, a concept-based neuro-symbolic architecture where concepts are represented as Signal Temporal Logic (STL) formulae. Leveraging kernel-based methods for STL, concept embeddings are learnt in an unsupervised manner through a cross-attention mechanism. The network makes class predictions through these concept embeddings, allowing for a meaningful explanation to be naturally extracted for each input. Our preliminary experiments with a simple CPS-based dataset show that our model is able to achieve great classification performance while ensuring local interpretability., Comment: 14 pages, 8 figures, submitted to 18th International Conference on Neural-Symbolic Learning and Reasoning (NeSy 2024)

Time-series analysis plays a pivotal role across a range of critical applications, from finance to healthcare, which involves various tasks, such as forecasting and classification. To handle the inherent complexities of time-series data, such as high dimensionality and noise, traditional supervised learning methods first annotate extensive labels for time-series data in each task, which is very costly and impractical in real-world applications. In contrast, pre-trained foundation models offer a promising alternative by leveraging unlabeled data to capture general time series patterns, which can then be fine-tuned for specific tasks. However, existing approaches to pre-training such models typically suffer from high-bias and low-generality issues due to the use of predefined and rigid augmentation operations and domain-specific data training. To overcome these limitations, this paper introduces UniCL, a universal and scalable contrastive learning framework designed for pretraining time-series foundation models across cross-domain datasets. Specifically, we propose a unified and trainable time-series augmentation operation to generate pattern-preserved, diverse, and low-bias time-series data by leveraging spectral information. Besides, we introduce a scalable augmentation algorithm capable of handling datasets with varying lengths, facilitating cross-domain pretraining. Extensive experiments on two benchmark datasets across eleven domains validate the effectiveness of UniCL, demonstrating its high generalization on time-series analysis across various fields.

Crowd counting and localization have become increasingly important in computer vision due to their wide-ranging applications. While point-based strategies have been widely used in crowd counting methods, they face a significant challenge, i.e., the lack of an effective learning strategy to guide the matching process. This deficiency leads to instability in matching point proposals to target points, adversely affecting overall performance. To address this issue, we introduce an effective approach to stabilize the proposal-target matching in point-based methods. We propose Auxiliary Point Guidance (APG) to provide clear and effective guidance for proposal selection and optimization, addressing the core issue of matching uncertainty. Additionally, we develop Implicit Feature Interpolation (IFI) to enable adaptive feature extraction in diverse crowd scenarios, further enhancing the model's robustness and accuracy. Extensive experiments demonstrate the effectiveness of our approach, showing significant improvements in crowd counting and localization performance, particularly under challenging conditions. The source codes and trained models will be made publicly available.

Experimental exploration of high-cost systems with safety constraints, common in engineering applications, is a challenging endeavor. Data-driven models offer a promising solution, but acquiring the requisite data remains expensive and is potentially unsafe. Safe active learning techniques prove essential, enabling the learning of high-quality models with minimal expensive data points and high safety. This paper introduces a safe active learning framework tailored for time-varying systems, addressing drift, seasonal changes, and complexities due to dynamic behavior. The proposed Time-aware Integrated Mean Squared Prediction Error (T-IMSPE) method minimizes posterior variance over current and future states, optimizing information gathering also in the time domain. Empirical results highlight T-IMSPE's advantages in model quality through toy and real-world examples. State of the art Gaussian processes are compatible with T-IMSPE. Our theoretical contributions include a clear delineation which Gaussian process kernels, domains, and weighting measures are suitable for T-IMSPE and even beyond for its non-time aware predecessor IMSPE.

In cardiac Magnetic Resonance Imaging (MRI) analysis, simultaneous myocardial segmentation and T2 quantification are crucial for assessing myocardial pathologies. Existing methods often address these tasks separately, limiting their synergistic potential. To address this, we propose SQNet, a dual-task network integrating Transformer and Convolutional Neural Network (CNN) components. SQNet features a T2-refine fusion decoder for quantitative analysis, leveraging global features from the Transformer, and a segmentation decoder with multiple local region supervision for enhanced accuracy. A tight coupling module aligns and fuses CNN and Transformer branch features, enabling SQNet to focus on myocardium regions. Evaluation on healthy controls (HC) and acute myocardial infarction patients (AMI) demonstrates superior segmentation dice scores (89.3/89.2) compared to state-of-the-art methods (87.7/87.9). T2 quantification yields strong linear correlations (Pearson coefficients: 0.84/0.93) with label values for HC/AMI, indicating accurate mapping. Radiologist evaluations confirm SQNet's superior image quality scores (4.60/4.58 for segmentation, 4.32/4.42 for T2 quantification) over state-of-the-art methods (4.50/4.44 for segmentation, 3.59/4.37 for T2 quantification). SQNet thus offers accurate simultaneous segmentation and quantification, enhancing cardiac disease diagnosis, such as AMI., Comment: 10 pages, 8 figures, 6 tables

In light of recent breakthroughs in large language models (LLMs) that have revolutionized natural language processing (NLP), there is an urgent need for new benchmarks to keep pace with the fast development of LLMs. In this paper, we propose CFLUE, the Chinese Financial Language Understanding Evaluation benchmark, designed to assess the capability of LLMs across various dimensions. Specifically, CFLUE provides datasets tailored for both knowledge assessment and application assessment. In knowledge assessment, it consists of 38K+ multiple-choice questions with associated solution explanations. These questions serve dual purposes: answer prediction and question reasoning. In application assessment, CFLUE features 16K+ test instances across distinct groups of NLP tasks such as text classification, machine translation, relation extraction, reading comprehension, and text generation. Upon CFLUE, we conduct a thorough evaluation of representative LLMs. The results reveal that only GPT-4 and GPT-4-turbo achieve an accuracy exceeding 60\% in answer prediction for knowledge assessment, suggesting that there is still substantial room for improvement in current LLMs. In application assessment, although GPT-4 and GPT-4-turbo are the top two performers, their considerable advantage over lightweight LLMs is noticeably diminished. The datasets and scripts associated with CFLUE are openly accessible at https://github.com/aliyun/cflue., Comment: Accepted by ACL 2024

Energy storage devices, such as batteries, thermal energy storages, and hydrogen systems, can help mitigate climate change by ensuring a more stable and sustainable power supply. To maximize the effectiveness of such energy storage, determining the appropriate charging and discharging amounts for each time period is crucial. Reinforcement learning is preferred over traditional optimization for the control of energy storage due to its ability to adapt to dynamic and complex environments. However, the continuous nature of charging and discharging levels in energy storage poses limitations for discrete reinforcement learning, and time-varying feasible charge-discharge range based on state of charge (SoC) variability also limits the conventional continuous reinforcement learning. In this paper, we propose a continuous reinforcement learning approach that takes into account the time-varying feasible charge-discharge range. An additional objective function was introduced for learning the feasible action range for each time period, supplementing the objectives of training the actor for policy learning and the critic for value learning. This actively promotes the utilization of energy storage by preventing them from getting stuck in suboptimal states, such as continuous full charging or discharging. This is achieved through the enforcement of the charging and discharging levels into the feasible action range. The experimental results demonstrated that the proposed method further maximized the effectiveness of energy storage by actively enhancing its utilization., Comment: ICLR 2024 Workshop: Tackling Climate Change with Machine Learning

Large language models have become increasingly prominent, also signaling a shift towards multimodality as the next frontier in artificial intelligence, where their embeddings are harnessed as prompts to generate textual content. Vision-language models (VLMs) stand at the forefront of this advancement, offering innovative ways to combine visual and textual data for enhanced understanding and interaction. However, this integration also enlarges the attack surface. Patch-based adversarial attack is considered the most realistic threat model in physical vision applications, as demonstrated in many existing literature. In this paper, we propose to address patched visual prompt injection, where adversaries exploit adversarial patches to generate target content in VLMs. Our investigation reveals that patched adversarial prompts exhibit sensitivity to pixel-wise randomization, a trait that remains robust even against adaptive attacks designed to counteract such defenses. Leveraging this insight, we introduce SmoothVLM, a defense mechanism rooted in smoothing techniques, specifically tailored to protect VLMs from the threat of patched visual prompt injectors. Our framework significantly lowers the attack success rate to a range between 0% and 5.0% on two leading VLMs, while achieving around 67.3% to 95.0% context recovery of the benign images, demonstrating a balance between security and usability., Comment: 15 pages

Remote sensing image-text retrieval constitutes a foundational aspect of remote sensing interpretation tasks, facilitating the alignment of vision and language representations. This paper introduces a prior instruction representation (PIR) learning paradigm that draws on prior knowledge to instruct adaptive learning of vision and text representations. Based on PIR, a domain-adapted remote sensing image-text retrieval framework PIR-ITR is designed to address semantic noise issues in vision-language understanding tasks. However, with massive additional data for pre-training the vision-language foundation model, remote sensing image-text retrieval is further developed into an open-domain retrieval task. Continuing with the above, we propose PIR-CLIP, a domain-specific CLIP-based framework for remote sensing image-text retrieval, to address semantic noise in remote sensing vision-language representations and further improve open-domain retrieval performance. In vision representation, Vision Instruction Representation (VIR) based on Spatial-PAE utilizes the prior-guided knowledge of the remote sensing scene recognition by building a belief matrix to select key features for reducing the impact of semantic noise. In text representation, Language Cycle Attention (LCA) based on Temporal-PAE uses the previous time step to cyclically activate the current time step to enhance text representation capability. A cluster-wise Affiliation Loss (AL) is proposed to constrain the inter-classes and to reduce the semantic confusion zones in the common subspace. Comprehensive experiments demonstrate that PIR could enhance vision and text representations and outperform the state-of-the-art methods of closed-domain and open-domain retrieval on two benchmark datasets, RSICD and RSITMD., Comment: 15 pages, 9 figures

In autonomous driving, accurately interpreting the movements of other road users and leveraging this knowledge to forecast future trajectories is crucial. This is typically achieved through the integration of map data and tracked trajectories of various agents. Numerous methodologies combine this information into a singular embedding for each agent, which is then utilized to predict future behavior. However, these approaches have a notable drawback in that they may lose exact location information during the encoding process. The encoding still includes general map information. However, the generation of valid and consistent trajectories is not guaranteed. This can cause the predicted trajectories to stray from the actual lanes. This paper introduces a new refinement module designed to project the predicted trajectories back onto the actual map, rectifying these discrepancies and leading towards more consistent predictions. This versatile module can be readily incorporated into a wide range of architectures. Additionally, we propose a novel scene encoder that handles all relations between agents and their environment in a single unified heterogeneous graph attention network. By analyzing the attention values on the different edges in this graph, we can gain unique insights into the neural network's inner workings leading towards a more explainable prediction.

The emergence of large language models (LLMs) capable of generating realistic texts and images has sparked ethical concerns across various sectors. In response, researchers in academia and industry are actively exploring methods to distinguish AI-generated content from human-authored material. However, a crucial question remains: What are the unique characteristics of AI-generated text? Addressing this gap, this study proposes StyloAI, a data-driven model that uses 31 stylometric features to identify AI-generated texts by applying a Random Forest classifier on two multi-domain datasets. StyloAI achieves accuracy rates of 81% and 98% on the test set of the AuTextification dataset and the Education dataset, respectively. This approach surpasses the performance of existing state-of-the-art models and provides valuable insights into the differences between AI-generated and human-authored texts., Comment: 25th International Conference on Artificial on Artificial Intelligence in Education(AIED 2024)

Most language models currently available are prone to self-contradiction during dialogues. To mitigate this issue, this study explores a novel contradictory dialogue processing task that aims to detect and modify contradictory statements in a conversation. This task is inspired by research on context faithfulness and dialogue comprehension, which have demonstrated that the detection and understanding of contradictions often necessitate detailed explanations. We develop a dataset comprising contradictory dialogues, in which one side of the conversation contradicts itself. Each dialogue is accompanied by an explanatory label that highlights the location and details of the contradiction. With this dataset, we present a Red Teaming framework for contradictory dialogue processing. The framework detects and attempts to explain the dialogue, then modifies the existing contradictory content using the explanation. Our experiments demonstrate that the framework improves the ability to detect contradictory dialogues and provides valid explanations. Additionally, it showcases distinct capabilities for modifying such dialogues. Our study highlights the importance of the logical inconsistency problem in conversational AI., Comment: 18 pages, 5 figures

This paper introduces a novel approach to predicting periodic time series using reservoir computing. The model is tailored to deliver precise forecasts of rhythms, a crucial aspect for tasks such as generating musical rhythm. Leveraging reservoir computing, our proposed method is ultimately oriented towards predicting human perception of rhythm. Our network accurately predicts rhythmic signals within the human frequency perception range. The model architecture incorporates primary and intermediate neurons tasked with capturing and transmitting rhythmic information. Two parameter matrices, denoted as c and k, regulate the reservoir's overall dynamics. We propose a loss function to adapt c post-training and introduce a dynamic selection (DS) mechanism that adjusts $k$ to focus on areas with outstanding contributions. Experimental results on a diverse test set showcase accurate predictions, further improved through real-time tuning of the reservoir via c and k. Comparative assessments highlight its superior performance compared to conventional models.

We introduce LatentTimePFN (LaT-PFN), a foundational Time Series model with a strong embedding space that enables zero-shot forecasting. To achieve this, we perform in-context learning in latent space utilizing a novel integration of the Prior-data Fitted Networks (PFN) and Joint Embedding Predictive Architecture (JEPA) frameworks. We leverage the JEPA framework to create a prediction-optimized latent representation of the underlying stochastic process that generates time series and combines it with contextual learning, using a PFN. Furthermore, we improve on preceding works by utilizing related time series as a context and introducing an abstract time axis. This drastically reduces training time and increases the versatility of the model by allowing any time granularity and forecast horizon. We show that this results in superior zero-shot predictions compared to established baselines. We also demonstrate our latent space produces informative embeddings of both individual time steps and fixed-length summaries of entire series. Finally, we observe the emergence of multi-step patch embeddings without explicit training, suggesting the model actively learns discrete tokens that encode local structures in the data, analogous to vision transformers., Comment: 9 pages plus references and appendix, 2 tables, 11 figures

The Learning-to-match (LTM) framework proves to be an effective inverse optimal transport approach for learning the underlying ground metric between two sources of data, facilitating subsequent matching. However, the conventional LTM framework faces scalability challenges, necessitating the use of the entire dataset each time the parameters of the ground metric are updated. In adapting LTM to the deep learning context, we introduce the mini-batch Learning-to-match (m-LTM) framework for audio-text retrieval problems. This framework leverages mini-batch subsampling and Mahalanobis-enhanced family of ground metrics. Moreover, to cope with misaligned training data in practice, we propose a variant using partial optimal transport to mitigate the harm of misaligned data pairs in training data. We conduct extensive experiments on audio-text matching problems using three datasets: AudioCaps, Clotho, and ESC-50. Results demonstrate that our proposed method is capable of learning rich and expressive joint embedding space, which achieves SOTA performance. Beyond this, the proposed m-LTM framework is able to close the modality gap across audio and text embedding, which surpasses both triplet and contrastive loss in the zero-shot sound event detection task on the ESC-50 dataset. Notably, our strategy of employing partial optimal transport with m-LTM demonstrates greater noise tolerance than contrastive loss, especially under varying noise ratios in training data on the AudioCaps dataset. Our code is available at https://github.com/v-manhlt3/m-LTM-Audio-Text-Retrieval