分享
Humans can retain old knowledge while learning new information, but Large Language Models (LLMs) often suffer from catastrophic forgetting when post-pretrained or supervised fine-tuned (SFT) on domain-specific data. Moreover, for Multimodal Large Language Models (MLLMs) which are composed of the LLM base and visual projector (e.g. LLaVA), a significant decline in performance on language benchmarks was observed compared to their single-modality counterparts. To address these challenges, we introduce a novel model-agnostic self-decompression method, Tree Generation (TG), that decompresses knowledge within LLMs into the training corpus. This paper focuses on TG-SFT, which can synthetically generate SFT data for the instruction tuning steps. By incorporating the dumped corpus during SFT for MLLMs, we significantly reduce the forgetting problem.
This paper presents the design and evaluation of a novel multi-level LLM interface for supermarket robots to assist customers. The proposed interface allows customers to convey their needs through both generic and specific queries. While state-of-the-art systems like OpenAI's GPTs are highly adaptable and easy to build and deploy, they still face challenges such as increased response times and limitations in strategic control of the underlying model for tailored use-case and cost optimization. Driven by the goal of developing faster and more efficient conversational agents, this paper advocates for using multiple smaller, specialized LLMs fine-tuned to handle different user queries based on their specificity and user intent. We compare this approach to a specialized GPT model powered by GPT-4 Turbo, using the Artificial Social Agent Questionnaire (ASAQ) and qualitative participant feedback in a counterbalanced within-subjects experiment. Our findings show that our multi-LLM chatbot architecture outperformed the benchmarked GPT model across all 13 measured criteria, with statistically significant improvements in four key areas: performance, user satisfaction, user-agent partnership, and self-image enhancement. The paper also presents a method for supermarket robot navigation by mapping the final chatbot response to correct shelf numbers, enabling the robot to sequentially navigate towards the respective products, after which lower-level robot perception, control, and planning can be used for automated object retrieval. We hope this work encourages more efforts into using multiple, specialized smaller models instead of relying on a single powerful, but more expensive and slower model.
Speech recognition is an essential start ring of human-computer interaction, and recently, deep learning models have achieved excellent success in this task. However, when the model training and private data provider are always separated, some security threats that make deep neural networks (DNNs) abnormal deserve to be researched. In recent years, the typical backdoor attacks have been researched in speech recognition systems. The existing backdoor methods are based on data poisoning. The attacker adds some incorporated changes to benign speech spectrograms or changes the speech components, such as pitch and timbre. As a result, the poisoned data can be detected by human hearing or automatic deep algorithms. To improve the stealthiness of data poisoning, we propose a non-neural and fast algorithm called Random Spectrogram Rhythm Transformation (RSRT) in this paper. The algorithm combines four steps to generate stealthy poisoned utterances. From the perspective of rhythm component transformation, our proposed trigger stretches or squeezes the mel spectrograms and recovers them back to signals. The operation keeps timbre and content unchanged for good stealthiness. Our experiments are conducted on two kinds of speech recognition tasks, including testing the stealthiness of poisoned samples by speaker verification and automatic speech recognition. The results show that our method has excellent effectiveness and stealthiness. The rhythm trigger needs a low poisoning rate and gets a very high attack success rate.
The process of identifying human emotion and affective states from speech is known as speech emotion recognition (SER). This is based on the observation that tone and pitch in the voice frequently convey underlying emotion. Speech recognition includes the ability to recognize emotions, which is becoming increasingly popular and in high demand. With the help of appropriate factors (such modalities, emotions, intensities, repetitions, etc.) found in the data, my research seeks to use the Convolutional Neural Network (CNN) to distinguish emotions from audio recordings and label them in accordance with the range of different emotions. I have developed a machine learning model to identify emotions from supplied audio files with the aid of machine learning methods. The evaluation is mostly focused on precision, recall, and F1 score, which are common machine learning metrics. To properly set up and train the machine learning framework, the main objective is to investigate the influence and cross-relation of all input and output parameters. To improve the ability to recognize intentions, a key condition for communication, I have evaluated emotions using my specialized machine learning algorithm via voice that would address the emotional state from voice with the help of digital healthcare, bridging the gap between human and artificial intelligence (AI).
Objective: To enhance the efficiency and accuracy of information retrieval from pharmacovigilance (PV) databases by employing Large Language Models (LLMs) to convert natural language queries (NLQs) into Structured Query Language (SQL) queries, leveraging a business context document. Materials and Methods: We utilized OpenAI's GPT-4 model within a retrieval-augmented generation (RAG) framework, enriched with a business context document, to transform NLQs into syntactically precise SQL queries. Each NLQ was presented to the LLM randomly and independently to prevent memorization. The study was conducted in three phases, varying query complexity, and assessing the LLM's performance both with and without the business context document. Results: Our approach significantly improved NLQ-to-SQL accuracy, increasing from 8.3\% with the database schema alone to 78.3\% with the business context document. This enhancement was consistent across low, medium, and high complexity queries, indicating the critical role of contextual knowledge in query generation. Discussion: The integration of a business context document markedly improved the LLM's ability to generate accurate and contextually relevant SQL queries. Performance achieved a maximum of 85\% when high complexity queries are excluded, suggesting promise for routine deployment. Conclusion: This study presents a novel approach to employing LLMs for safety data retrieval and analysis, demonstrating significant advancements in query generation accuracy. The methodology offers a framework applicable to various data-intensive domains, enhancing the accessibility and efficiency of information retrieval for non-technical users.
Federated learning (FL) enables multiple parties to collaboratively fine-tune an large language model (LLM) without the need of direct data sharing. Ideally, by training on decentralized data that is aligned with human preferences and safety principles, federated instruction tuning can result in an LLM that could behave in a helpful and safe manner. In this paper, we for the first time reveal the vulnerability of safety alignment in FedIT by proposing a simple, stealthy, yet effective safety attack method. Specifically, the malicious clients could automatically generate attack data without involving manual efforts and attack the FedIT system by training their local LLMs on such attack data. Unfortunately, this proposed safety attack not only can compromise the safety alignment of LLM trained via FedIT, but also can not be effectively defended against by many existing FL defense methods. Targeting this, we further propose a post-hoc defense method, which could rely on a fully automated pipeline: generation of defense data and further fine-tuning of the LLM. Extensive experiments show that our safety attack method can significantly compromise the LLM's safety alignment (e.g., reduce safety rate by 70\%), which can not be effectively defended by existing defense methods (at most 4\% absolute improvement), while our safety defense method can significantly enhance the attacked LLM's safety alignment (at most 69\% absolute improvement).
Research in 3D mapping is crucial for smart city applications, yet the cost of acquiring 3D data often hinders progress. Visual localization, particularly monocular camera position estimation, offers a solution by determining the camera's pose solely through visual cues. However, this task is challenging due to limited data from a single camera. To tackle these challenges, we organized the AISG-SLA Visual Localization Challenge (VLC) at IJCAI 2023 to explore how AI can accurately extract camera pose data from 2D images in 3D space. The challenge attracted over 300 participants worldwide, forming 50+ teams. Winning teams achieved high accuracy in pose estimation using images from a car-mounted camera with low frame rates. The VLC dataset is available for research purposes upon request via vlc-dataset@aisingapore.org.
Large Language Models (LLM) alignment aims to prevent models from producing content that misaligns with human expectations, which can lead to ethical and legal concerns. In the last few years, Reinforcement Learning from Human Feedback (RLHF) has been the most prominent method for achieving alignment. Due to challenges in stability and scalability with RLHF stages, which arise from the complex interactions between multiple models, researchers are exploring alternative methods to achieve effects comparable to those of RLHF. However, these methods often rely on large high-quality datasets. Despite some methods considering the generation of additional data to expand datasets, they often treat model training and data generation as separate and static processes, overlooking the fact that these processes are highly interdependent, leading to inefficient utilization of the generated data. To deal with this problem, we propose PLE, i.e., Progressively Label Enhancement for LLM Alignment, a framework that dynamically adjusts the model's training process based on the evolving quality of the generated data. Specifically, we prompt the model to generate responses for both the original query and the query guided by a set of carefully designed principles, and then utilize a dynamic threshold to determine the appropriate training approach for both responses based on their corresponding reward scores. Experimental results demonstrate the effectiveness of PLE compared to existing LLM alignment methods.
Sarcasm is a type of irony, characterized by an inherent mismatch between the literal interpretation and the intended connotation. Though sarcasm detection in text has been extensively studied, there are situations in which textual input alone might be insufficient to perceive sarcasm. The inclusion of additional contextual cues, such as images, is essential to recognize sarcasm in social media data effectively. This study presents a novel framework for multimodal sarcasm detection that can process input triplets. Two components of these triplets comprise the input text and its associated image, as provided in the datasets. Additionally, a supplementary modality is introduced in the form of descriptive image captions. The motivation behind incorporating this visual semantic representation is to more accurately capture the discrepancies between the textual and visual content, which are fundamental to the sarcasm detection task. The primary contributions of this study are: (1) a robust textual feature extraction branch that utilizes a cross-lingual language model; (2) a visual feature extraction branch that incorporates a self-regulated residual ConvNet integrated with a lightweight spatially aware attention module; (3) an additional modality in the form of image captions generated using an encoder-decoder architecture capable of reading text embedded in images; (4) distinct attention modules to effectively identify the incongruities between the text and two levels of image representations; (5) multi-level cross-domain semantic incongruity representation achieved through feature fusion. Compared with cutting-edge baselines, the proposed model achieves the best accuracy of 92.89% and 64.48%, respectively, on the Twitter multimodal sarcasm and MultiBully datasets.
Despite their impressive performance, large language models (LLMs) such as ChatGPT are known to pose important risks. One such set of risks arises from misplaced confidence, whether over-confidence or under-confidence, that the models have in their inference. While the former is well studied, the latter is not, leading to an asymmetry in understanding the comprehensive risk of the model based on misplaced confidence. In this paper, we address this asymmetry by defining two types of risk (decision and composite risk), and proposing an experimental framework consisting of a two-level inference architecture and appropriate metrics for measuring such risks in both discriminative and generative LLMs. The first level relies on a decision rule that determines whether the underlying language model should abstain from inference. The second level (which applies if the model does not abstain) is the model's inference. Detailed experiments on four natural language commonsense reasoning datasets using both an open-source ensemble-based RoBERTa model and ChatGPT, demonstrate the practical utility of the evaluation framework. For example, our results show that our framework can get an LLM to confidently respond to an extra 20.1% of low-risk inference tasks that other methods might misclassify as high-risk, and skip 19.8% of high-risk tasks, which would have been answered incorrectly.
This work presents an audio-visual interactive chatbot (AVIN-Chat) system that allows users to have face-to-face conversations with 3D avatars in real-time. Compared to the previous chatbot services, which provide text-only or speech-only communications, the proposed AVIN-Chat can offer audio-visual communications providing users with a superior experience quality. In addition, the proposed AVIN-Chat emotionally speaks and expresses according to the user's emotional state. Thus, it enables users to establish a strong bond with the chatbot system, increasing the user's immersion. Through user subjective tests, it is demonstrated that the proposed system provides users with a higher sense of immersion than previous chatbot systems. The demonstration video is available at https://www.youtube.com/watch?v=Z74uIV9k7_k.
Verbalization of robot experience, i.e., summarization of and question answering about a robot's past, is a crucial ability for improving human-robot interaction. Previous works applied rule-based systems or fine-tuned deep models to verbalize short (several-minute-long) streams of episodic data, limiting generalization and transferability. In our work, we apply large pretrained models to tackle this task with zero or few examples, and specifically focus on verbalizing life-long experiences. For this, we derive a tree-like data structure from episodic memory (EM), with lower levels representing raw perception and proprioception data, and higher levels abstracting events to natural language concepts. Given such a hierarchical representation built from the experience stream, we apply a large language model as an agent to interactively search the EM given a user's query, dynamically expanding (initially collapsed) tree nodes to find the relevant information. The approach keeps computational costs low even when scaling to months of robot experience data. We evaluate our method on simulated household robot data, human egocentric videos, and real-world robot recordings, demonstrating its flexibility and scalability.