Recent advancements in large language models (LLMs) have revolutionized various domains, bringing significant progress and new opportunities. Despite progress in speech-related tasks, LLMs have not been sufficiently explored in multi-talker scenarios. In this work, we present a pioneering effort to investigate the capability of LLMs in transcribing speech in multi-talker environments, following versatile instructions related to multi-talker automatic speech recognition (ASR), target talker ASR, and ASR based on specific talker attributes such as sex, occurrence order, language, and keyword spoken. Our approach utilizes WavLM and Whisper encoder to extract multi-faceted speech representations that are sensitive to speaker characteristics and semantic context. These representations are then fed into an LLM fine-tuned using LoRA, enabling the capabilities for speech comprehension and transcription. Comprehensive experiments reveal the promising performance of our proposed system, MT-LLM, in cocktail party scenarios, highlighting the potential of LLM to handle speech-related tasks based on user instructions in such complex settings. The code, model, and samples are available at https://github.com/cuhealthybrains/MT-LLM.
Large language models (LLMs) have demonstrated remarkable capabilities in tasks requiring reasoning and multi-step problem-solving through the use of chain-of-thought (CoT) prompting. However, generating the full CoT process results in significantly longer output sequences, leading to increased computational costs and latency during inference. To address this challenge, we propose a novel approach to compress the CoT process through semantic alignment, enabling more efficient decoding while preserving the benefits of CoT reasoning. Our method introduces an auxiliary CoT model that learns to generate and compress the full thought process into a compact special token representation semantically aligned with the original CoT output. This compressed representation is then integrated into the input of the Hidden Chain-of-Thought (HCoT) model. The training process follows a two-stage procedure: First, the CoT model is optimized to generate the compressed token representations aligned with the ground-truth CoT outputs using a contrastive loss. Subsequently, with the CoT model parameters frozen, the HCoT model is fine-tuned to generate accurate subsequent predictions conditioned on the prefix instruction and the compressed CoT representations from the CoT model. Extensive experiments across three challenging domains - mathematical reasoning, agent invocation, and question answering - demonstrate that our semantic compression approach achieves competitive or improved performance compared to the full CoT baseline, while providing significant speedups of at least 1.5x in decoding time. Moreover, incorporating contrastive learning objectives further enhances the quality of the compressed representations, leading to better CoT prompting and improved task accuracy. Our work paves the way for more efficient exploitation of multi-step reasoning capabilities in LLMs across a wide range of applications.
Recommender Systems (RS) play a pivotal role in boosting user satisfaction by providing personalized product suggestions in domains such as e-commerce and entertainment. This study examines the integration of multimodal data text and audio into large language models (LLMs) with the aim of enhancing recommendation performance. Traditional text and audio recommenders encounter limitations such as the cold-start problem, and recent advancements in LLMs, while promising, are computationally expensive. To address these issues, Low-Rank Adaptation (LoRA) is introduced, which enhances efficiency without compromising performance. The ATFLRec framework is proposed to integrate audio and text modalities into a multimodal recommendation system, utilizing various LoRA configurations and modality fusion techniques. Results indicate that ATFLRec outperforms baseline models, including traditional and graph neural network-based approaches, achieving higher AUC scores. Furthermore, separate fine-tuning of audio and text data with distinct LoRA modules yields optimal performance, with different pooling methods and Mel filter bank numbers significantly impacting performance. This research offers valuable insights into optimizing multimodal recommender systems and advancing the integration of diverse data modalities in LLMs.
To make sense of massive data, we often fit simplified models and then interpret the parameters; for example, we cluster the text embeddings and then interpret the mean parameters of each cluster. However, these parameters are often high-dimensional and hard to interpret. To make model parameters directly interpretable, we introduce a family of statistical models -- including clustering, time series, and classification models -- parameterized by natural language predicates. For example, a cluster of text about COVID could be parameterized by the predicate "discusses COVID". To learn these statistical models effectively, we develop a model-agnostic algorithm that optimizes continuous relaxations of predicate parameters with gradient descent and discretizes them by prompting language models (LMs). Finally, we apply our framework to a wide range of problems: taxonomizing user chat dialogues, characterizing how they evolve across time, finding categories where one language model is better than the other, clustering math problems based on subareas, and explaining visual features in memorable images. Our framework is highly versatile, applicable to both textual and visual domains, can be easily steered to focus on specific properties (e.g. subareas), and explains sophisticated concepts that classical methods (e.g. n-gram analysis) struggle to produce.
Knowledge tagging for questions is vital in modern intelligent educational applications, including learning progress diagnosis, practice question recommendations, and course content organization. Traditionally, these annotations have been performed by pedagogical experts, as the task demands not only a deep semantic understanding of question stems and knowledge definitions but also a strong ability to link problem-solving logic with relevant knowledge concepts. With the advent of advanced natural language processing (NLP) algorithms, such as pre-trained language models and large language models (LLMs), pioneering studies have explored automating the knowledge tagging process using various machine learning models. In this paper, we investigate the use of a multi-agent system to address the limitations of previous algorithms, particularly in handling complex cases involving intricate knowledge definitions and strict numerical constraints. By demonstrating its superior performance on the publicly available math question knowledge tagging dataset, MathKnowCT, we highlight the significant potential of an LLM-based multi-agent system in overcoming the challenges that previous methods have encountered. Finally, through an in-depth discussion of the implications of automating knowledge tagging, we underscore the promising results of deploying LLM-based algorithms in educational contexts.
Scholarly communication is a rapid growing field containing a wealth of knowledge. However, due to its unstructured and document format, it is challenging to extract useful information from them through conventional document retrieval methods. Scholarly knowledge graphs solve this problem, by representing the documents in a semantic network, providing, hidden insights, summaries and ease of accessibility through queries. Naturally, question answering for scholarly graphs expands the accessibility to a wider audience. But some of the knowledge in this domain is still presented as unstructured text, thus requiring a hybrid solution for question answering systems. In this paper, we present a two step solution using open source Large Language Model(LLM): Llama3.1 for Scholarly-QALD dataset. Firstly, we extract the context pertaining to the question from different structured and unstructured data sources: DBLP, SemOpenAlex knowledge graphs and Wikipedia text. Secondly, we implement prompt engineering to improve the information retrieval performance of the LLM. Our approach achieved an F1 score of 40% and also observed some anomalous responses from the LLM, that are discussed in the final part of the paper.
Large multimodal foundation models have been extensively utilized for image recognition tasks guided by instructions, yet there remains a scarcity of domain expertise in industrial vibration signal analysis. This paper presents a pipeline named VSLLaVA that leverages a large language model to integrate expert knowledge for identification of signal parameters and diagnosis of faults. Within this pipeline, we first introduce an expert rule-assisted signal generator. The generator merges signal provided by vibration analysis experts with domain-specific parameter identification and fault diagnosis question-answer pairs to build signal-question-answer triplets. Then we use these triplets to apply low-rank adaptation methods for fine-tuning the linear layers of the Contrastive Language-Image Pretraining (CLIP) and large language model, injecting multimodal signal processing knowledge. Finally, the fine-tuned model is assessed through the combined efforts of large language model and expert rules to evaluate answer accuracy and relevance, which showcases enhanced performance in identifying, analyzing various signal parameters, and diagnosing faults. These enhancements indicate the potential of this pipeline to build a foundational model for future industrial signal analysis and monitoring.
Multimodal language modeling has enabled breakthroughs for representation learning, yet remains unexplored in the realm of functional brain data for pathology detection. This paper pioneers EEG-language models (ELMs) trained on clinical reports and 15000 EEGs. We propose to combine multimodal alignment in this novel domain with timeseries cropping and text segmentation, enabling an extension based on multiple instance learning to alleviate misalignment between irrelevant EEG or text segments. Our multimodal models significantly improve pathology detection compared to EEG-only models across four evaluations and for the first time enable zero-shot classification as well as retrieval of both neural signals and reports. In sum, these results highlight the potential of ELMs, representing significant progress for clinical applications.
The adaptive immune response, largely mediated by B-cell receptors (BCRs), plays a crucial role for effective pathogen neutralization due to its diversity and antigen specificity. Designing BCRs de novo, or from scratch, has been challenging because of their complex structure and diverse binding requirements. Protein language models (PLMs) have shown remarkable performance in contextualizing and performing various downstream tasks without relying on structural information. However, these models often lack a comprehensive understanding of the entire protein space, which limits their application in antibody design. In this study, we introduce Antibody Generative Pretrained Transformer (AbGPT), a model fine-tuned from a foundational PLM to enable a more informed design of BCR sequences. Using a custom generation and filtering pipeline, AbGPT successfully generated a high-quality library of 15,000 BCR sequences, demonstrating a strong understanding of the intrinsic variability and conserved regions within the antibody repertoire.
Speech emotion recognition (SER), the task of identifying the expression of emotion from spoken content, is challenging due to the difficulty in extracting representations that capture emotional attributes from speech. The scarcity of labeled datasets further complicates the challenge where large models are prone to over-fitting. In this paper, we propose CARE (Content and Acoustic Representations of Emotions), where we design a dual encoding scheme which emphasizes semantic and acoustic factors of speech. While the semantic encoder is trained using distillation from utterance-level text representations, the acoustic encoder is trained to predict low-level frame-wise features of the speech signal. The proposed dual encoding scheme is a base-sized model trained only on unsupervised raw speech. With a simple light-weight classification model trained on the downstream task, we show that the CARE embeddings provide effective emotion recognition on a variety of datasets. We compare the proposal with several other self-supervised models as well as recent large-language model based approaches. In these evaluations, the proposed CARE is shown to be the best performing model based on average performance across 8 diverse datasets. We also conduct several ablation studies to analyze the importance of various design choices.
End-to-end transformer-based models epitomize the cutting-edge in Automatic Speech Recognition (ASR) systems. Despite their substantial benefits, these models demand extensive training data to perform optimally, presenting a significant challenge for low-resource languages such as Central Kurdish. Addressing this issue requires innovative methods and techniques. This paper aims to develop an ASR system for Intermediate Kurdish by collecting a robust corpus of speech, using the N-GRAM language model, and utilizing an external Kurdish tokenizer for refinement and integration techniques to enhance the model's performance. We collect a comprehensive 100-hour speech corpus from diverse sources. Additionally, applied fine-tuning techniques to our speech corpus on Persian, English, and Arabic pre-trained models, specifically utilizing the xls-r-300m, xls-r-1b, and xls-r-2b Wav2vec 2.0 models. And utilized language models trained by 3-gram and 4-gram from a large text corpus of 300 million tokens. The fine-tuned xls-r-2b model, combined with a 3-gram language model and included external Kurdish tokenizer, achieved the best performance, yielding a Word Error Rate (WER) of 10.0% on the validation set and 11.8% on the Asosoft test set. The ASR model has demonstrated the advantages of having a large vocabulary compared to the existing Kurdish ASR models. Compared to other models, it produced more accurate and higher performance outcomes by working with a lower error rate.
Our modern life has grown to depend on many and nearly ubiquitous large complex engineering systems. Many disciplines now seemingly ask the same question: ``In the face of assumed disruption, to what degree will these systems continue to perform and when will they be able to bounce back to normal operation"? Furthermore, there is a growing recognition that the greatest societal challenges of the Anthropocene era are intertwined, necessitating a convergent systems-of-systems modeling and analysis framework based upon reconciled ontologies, data, and theoretical methods. Consequently, this paper develops a methodology for hetero-functional graph resilience analysis and demonstrates it on a convergent system-of-systems. It uses the Systems Modeling Language, model-based systems engineering and Hetero-Functional Graph Theory (HFGT) to overcome the convergence research challenges when constructing models and measures from multiple disciplines for systems resilience. The paper includes both the ``survival" as well as ``recovery" components of resilience. It also strikes a middle ground between two disparate approaches to resilience measurement: structural measurement of formal graphs and detailed behavioral simulation. This paper also generalizes a previous resilience measure based on HFGT and benefits from recent theoretical and computational developments in HFGT. To demonstrate the methodological developments, the resilience analysis is conducted on a hypothetical energy-water nexus system of moderate size as a type of system-of-systems.
Large Language Models (LLMs) have demonstrated substantial potential for error correction in Automatic Speech Recognition (ASR). However, most research focuses on utterances from short-duration speech recordings, which are the predominant form of speech data for supervised ASR training. This paper investigates the effectiveness of LLMs for error correction in full-text generated by ASR systems from longer speech recordings, such as transcripts from podcasts, news broadcasts, and meetings. First, we develop a Chinese dataset for full-text error correction, named ChFT, utilizing a pipeline that involves text-to-speech synthesis, ASR, and error-correction pair extractor. This dataset enables us to correct errors across contexts, including both full-text and segment, and to address a broader range of error types, such as punctuation restoration and inverse text normalization, thus making the correction process comprehensive. Second, we fine-tune a pre-trained LLM on the constructed dataset using a diverse set of prompts and target formats, and evaluate its performance on full-text error correction. Specifically, we design prompts based on full-text and segment, considering various output formats, such as directly corrected text and JSON-based error-correction pairs. Through various test settings, including homogeneous, up-to-date, and hard test sets, we find that the fine-tuned LLMs perform well in the full-text setting with different prompts, each presenting its own strengths and weaknesses. This establishes a promising baseline for further research. The dataset is available on the website.
Stable pre-training is essential for achieving better-performing language models. However, tracking pre-training stability by calculating gradient variance at every step is impractical due to the significant computational costs. We explore Token Embedding Variability (TEV) as a simple and efficient proxy for assessing pre-training stability in language models with pre-layer normalization, given that shallower layers are more prone to gradient explosion (section 2.2). Moreover, we propose Multi-head Low-Rank Attention (MLRA) as an architecture to alleviate such instability by limiting the exponential growth of output embedding variance, thereby preventing the gradient explosion (section 3.2). Empirical results on GPT-2 with MLRA demonstrate increased stability and lower perplexity, particularly in deeper models.
Ethereum faces growing fraud threats. Current fraud detection methods, whether employing graph neural networks or sequence models, fail to consider the semantic information and similarity patterns within transactions. Moreover, these approaches do not leverage the potential synergistic benefits of combining both types of models. To address these challenges, we propose TLMG4Eth that combines a transaction language model with graph-based methods to capture semantic, similarity, and structural features of transaction data in Ethereum. We first propose a transaction language model that converts numerical transaction data into meaningful transaction sentences, enabling the model to learn explicit transaction semantics. Then, we propose a transaction attribute similarity graph to learn transaction similarity information, enabling us to capture intuitive insights into transaction anomalies. Additionally, we construct an account interaction graph to capture the structural information of the account transaction network. We employ a deep multi-head attention network to fuse transaction semantic and similarity embeddings, and ultimately propose a joint training approach for the multi-head attention network and the account interaction graph to obtain the synergistic benefits of both.
Generative models aim to simulate realistic effects of various actions across different contexts, from text generation to visual effects. Despite significant efforts to build real-world simulators, the application of generative models to virtual worlds, like financial markets, remains under-explored. In financial markets, generative models can simulate complex market effects of participants with various behaviors, enabling interaction under different market conditions, and training strategies without financial risk. This simulation relies on the finest structured data in financial market like orders thus building the finest realistic simulation. We propose Large Market Model (LMM), an order-level generative foundation model, for financial market simulation, akin to language modeling in the digital world. Our financial Market Simulation engine (MarS), powered by LMM, addresses the domain-specific need for realistic, interactive and controllable order generation. Key observations include LMM's strong scalability across data size and model complexity, and MarS's robust and practicable realism in controlled generation with market impact. We showcase MarS as a forecast tool, detection system, analysis platform, and agent training environment, thus demonstrating MarS's "paradigm shift" potential for a variety of financial applications. We release the code of MarS at https://github.com/microsoft/MarS/.
Embeddings-as-a-Service (EaaS) is a service offered by large language model (LLM) developers to supply embeddings generated by LLMs. Previous research suggests that EaaS is prone to imitation attacks -- attacks that clone the underlying EaaS model by training another model on the queried embeddings. As a result, EaaS watermarks are introduced to protect the intellectual property of EaaS providers. In this paper, we first show that existing EaaS watermarks can be removed by paraphrasing when attackers clone the model. Subsequently, we propose a novel watermarking technique that involves linearly transforming the embeddings, and show that it is empirically and theoretically robust against paraphrasing.
We present a new perspective on how readers integrate context during real-time language comprehension. Our proposals build on surprisal theory, which posits that the processing effort of a linguistic unit (e.g., a word) is an affine function of its in-context information content. We first observe that surprisal is only one out of many potential ways that a contextual predictor can be derived from a language model. Another one is the pointwise mutual information (PMI) between a unit and its context, which turns out to yield the same predictive power as surprisal when controlling for unigram frequency. Moreover, both PMI and surprisal are correlated with frequency. This means that neither PMI nor surprisal contains information about context alone. In response to this, we propose a technique where we project surprisal onto the orthogonal complement of frequency, yielding a new contextual predictor that is uncorrelated with frequency. Our experiments show that the proportion of variance in reading times explained by context is a lot smaller when context is represented by the orthogonalized predictor. From an interpretability standpoint, this indicates that previous studies may have overstated the role that context has in predicting reading times.
People often capture memories through photos, screenshots, and videos. While existing AI-based tools enable querying this data using natural language, they only support retrieving individual pieces of information like certain objects in photos, and struggle with answering more complex queries that involve interpreting interconnected memories like sequential events. We conducted a one-month diary study to collect realistic user queries and generated a taxonomy of necessary contextual information for integrating with captured memories. We then introduce OmniQuery, a novel system that is able to answer complex personal memory-related questions that require extracting and inferring contextual information. OmniQuery augments individual captured memories through integrating scattered contextual information from multiple interconnected memories. Given a question, OmniQuery retrieves relevant augmented memories and uses a large language model (LLM) to generate answers with references. In human evaluations, we show the effectiveness of OmniQuery with an accuracy of 71.5%, outperforming a conventional RAG system by winning or tying for 74.5% of the time.
The rapid evolution of cyber threats necessitates innovative solutions for detecting and analyzing malicious activity. Honeypots, which are decoy systems designed to lure and interact with attackers, have emerged as a critical component in cybersecurity. In this paper, we present a novel approach to creating realistic and interactive honeypot systems using Large Language Models (LLMs). By fine-tuning a pre-trained open-source language model on a diverse dataset of attacker-generated commands and responses, we developed a honeypot capable of sophisticated engagement with attackers. Our methodology involved several key steps: data collection and processing, prompt engineering, model selection, and supervised fine-tuning to optimize the model's performance. Evaluation through similarity metrics and live deployment demonstrated that our approach effectively generates accurate and informative responses. The results highlight the potential of LLMs to revolutionize honeypot technology, providing cybersecurity professionals with a powerful tool to detect and analyze malicious activity, thereby enhancing overall security infrastructure.