Abstract:Recent advancements in deep learning have significantly revolutionized the field of clinical diagnosis and treatment, offering novel approaches to improve diagnostic precision and treatment efficacy across diverse clinical domains, thus driving the pursuit of precision medicine. The growing availability of multi-organ and multimodal datasets has accelerated the development of large-scale Medical Multimodal Foundation Models (MMFMs). These models, known for their strong generalization capabilities and rich representational power, are increasingly being adapted to address a wide range of clinical tasks, from early diagnosis to personalized treatment strategies. This review offers a comprehensive analysis of recent developments in MMFMs, focusing on three key aspects: datasets, model architectures, and clinical applications. We also explore the challenges and opportunities in optimizing multimodal representations and discuss how these advancements are shaping the future of healthcare by enabling improved patient outcomes and more efficient clinical workflows.
Abstract:This paper provides a comprehensive survey of the latest research on multilingual large language models (MLLMs). MLLMs not only are able to understand and generate language across linguistic boundaries, but also represent an important advancement in artificial intelligence. We first discuss the architecture and pre-training objectives of MLLMs, highlighting the key components and methodologies that contribute to their multilingual capabilities. We then discuss the construction of multilingual pre-training and alignment datasets, underscoring the importance of data quality and diversity in enhancing MLLM performance. An important focus of this survey is on the evaluation of MLLMs. We present a detailed taxonomy and roadmap covering the assessment of MLLMs' cross-lingual knowledge, reasoning, alignment with human values, safety, interpretability and specialized applications. Specifically, we extensively discuss multilingual evaluation benchmarks and datasets, and explore the use of LLMs themselves as multilingual evaluators. To enhance MLLMs from black to white boxes, we also address the interpretability of multilingual capabilities, cross-lingual transfer and language bias within these models. Finally, we provide a comprehensive review of real-world applications of MLLMs across diverse domains, including biology, medicine, computer science, mathematics and law. We showcase how these models have driven innovation and improvements in these specialized fields while also highlighting the challenges and opportunities in deploying MLLMs within diverse language communities and application scenarios. We listed the paper related in this survey and publicly available at https://github.com/tjunlp-lab/Awesome-Multilingual-LLMs-Papers.
Abstract:Dynamic Range Compression (DRC) is a popular audio effect used to control the dynamic range of a signal. Inverting DRC can also help to restore the original dynamics to produce new mixes and/or to improve the overall quality of the audio signal. Since, state-of-the-art DRC inversion techniques either ignore parameters or require precise parameters that are difficult to estimate, we fill the gap by combining a model-based approach with neural networks for DRC inversion. To this end, depending on the scenario, we use different neural networks to estimate DRC parameters. Then, a model-based inversion is completed to restore the original audio signal. Our experimental results show the effectiveness and robustness of the proposed method in comparison to several state-of-the-art methods, when applied on two music datasets.
Abstract:We assume accurate observation of battery state of charge (SOC) and precise speed curves when addressing the constrained optimal fuel consumption (COFC) problem via constrained reinforcement learning (CRL). However, in practice, SOC measurements are often distorted by noise or confidentiality protocols, and actual reference speeds may deviate from expectations. We aim to minimize fuel consumption while maintaining SOC balance under observational perturbations in SOC and speed. This work first worldwide uses seven training approaches to solve the COFC problem under five types of perturbations, including one based on a uniform distribution, one designed to maximize rewards, one aimed at maximizing costs, and one along with its improved version that seeks to decrease reward on Toyota Hybrid Systems (THS) under New European Driving Cycle (NEDC) condition. The result verifies that the six can successfully solve the COFC problem under observational perturbations, and we further compare the robustness and safety of these training approaches and analyze their impact on optimal fuel consumption.
Abstract:Are Multi-modal Large Language Models (MLLMs) stochastic parrots? Do they genuinely understand and are capable of performing the tasks they excel at? This paper aims to explore the fundamental basis of MLLMs, i.e. core cognitive abilities that human intelligence builds upon to perceive, comprehend, and reason. To this end, we propose CogDevelop2K, a comprehensive benchmark that spans 12 sub-concepts from fundamental knowledge like object permanence and boundary to advanced reasoning like intentionality understanding, structured via the developmental trajectory of a human mind. We evaluate 46 MLLMs on our benchmarks. Comprehensively, we further evaluate the influence of evaluation strategies and prompting techniques. Surprisingly, we observe a reversed cognitive developmental trajectory compared to humans.
Abstract:Reinforcement learning from human feedback (RLHF) has demonstrated effectiveness in aligning large language models (LLMs) with human preferences. However, token-level RLHF suffers from the credit assignment problem over long sequences, where delayed rewards make it challenging for the model to discern which actions contributed to successful outcomes. This hinders learning efficiency and slows convergence. In this paper, we propose MA-RLHF, a simple yet effective RLHF framework that incorporates macro actions -- sequences of tokens or higher-level language constructs -- into the learning process. By operating at this higher level of abstraction, our approach reduces the temporal distance between actions and rewards, facilitating faster and more accurate credit assignment. This results in more stable policy gradient estimates and enhances learning efficiency within each episode, all without increasing computational complexity during training or inference. We validate our approach through extensive experiments across various model sizes and tasks, including text summarization, dialogue generation, question answering, and program synthesis. Our method achieves substantial performance improvements over standard RLHF, with performance gains of up to 30% in text summarization and code generation, 18% in dialogue, and 8% in question answering tasks. Notably, our approach reaches parity with vanilla RLHF 1.7x to 2x faster in terms of training time and continues to outperform it with further training. We will make our code and data publicly available at https://github.com/ernie-research/MA-RLHF .
Abstract:Large language models (LLMs) have demonstrated prowess in a wide range of tasks. However, many LLMs exhibit significant performance discrepancies between high- and low-resource languages. To mitigate this challenge, we present FuxiTranyu, an open-source multilingual LLM, which is designed to satisfy the need of the research community for balanced and high-performing multilingual capabilities. FuxiTranyu-8B, the base model with 8 billion parameters, is trained from scratch on a meticulously balanced multilingual data repository that contains 600 billion tokens covering 43 natural languages and 16 programming languages. In addition to the base model, we also develop two instruction-tuned models: FuxiTranyu-8B-SFT that is fine-tuned on a diverse multilingual instruction dataset, and FuxiTranyu-8B-DPO that is further refined with DPO on a preference dataset for enhanced alignment ability. Extensive experiments on a wide range of multilingual benchmarks demonstrate the competitive performance of FuxiTranyu against existing multilingual LLMs, e.g., BLOOM-7B, PolyLM-13B, Llama-2-Chat-7B and Mistral-7B-Instruct. Interpretability analyses at both the neuron and representation level suggest that FuxiTranyu is able to learn consistent multilingual representations across different languages. To promote further research into multilingual LLMs and their working mechanisms, we release both the base and instruction-tuned FuxiTranyu models together with 58 pretraining checkpoints at HuggingFace and Github.
Abstract:Retrieving 3D bone anatomy from biplanar X-ray images is crucial since it can significantly reduce radiation exposure compared to traditional CT-based methods. Although various deep learning models have been proposed to address this complex task, they suffer from two limitations: 1) They employ voxel representation for bone shape and exploit 3D convolutional layers to capture anatomy prior, which are memory-intensive and limit the reconstruction resolution. 2) They overlook the prevalent occlusion effect within X-ray images and directly extract features using a simple loss, which struggles to fully exploit complex X-ray information. To tackle these concerns, we present Spatial-division Augmented Occupancy Field~(SdAOF). SdAOF adopts the continuous occupancy field for shape representation, reformulating the reconstruction problem as a per-point occupancy value prediction task. Its implicit and continuous nature enables memory-efficient training and fine-scale surface reconstruction at different resolutions during the inference. Moreover, we propose a novel spatial-division augmented distillation strategy to provide feature-level guidance for capturing the occlusion relationship. Extensive experiments on the pelvis reconstruction dataset show that SdAOF outperforms state-of-the-art methods and reconstructs fine-scale bone surfaces.The code is available at https://github.com/xmed-lab/SdAOF
Abstract:Fabric manipulation dynamically is commonly seen in manufacturing and domestic settings. While dynamically manipulating a fabric piece to reach a target state is highly efficient, this task presents considerable challenges due to the varying properties of different fabrics, complex dynamics when interacting with environments, and meeting required goal conditions. To address these challenges, we present \textit{One Fling to Goal}, an algorithm capable of handling fabric pieces with diverse shapes and physical properties across various scenarios. Our method learns a graph-based dynamics model equipped with environmental awareness. With this dynamics model, we devise a real-time controller to enable high-speed fabric manipulation in one attempt, requiring less than 3 seconds to finish the goal-conditioned task. We experimentally validate our method on a goal-conditioned manipulation task in five diverse scenarios. Our method significantly improves this goal-conditioned task, achieving an average error of 13.2mm in complex scenarios. Our method can be seamlessly transferred to real-world robotic systems and generalized to unseen scenarios in a zero-shot manner.
Abstract:In this paper, we present the findings of our Project ALPINE which stands for ``Autoregressive Learning for Planning In NEtworks." Project ALPINE initiates a theoretical investigation into the development of planning capabilities in Transformer-based language models through their autoregressive learning mechanisms, aiming to identify any potential limitations in their planning abilities. We abstract planning as a network path-finding task where the objective is to generate a valid path from a specified source node to a designated target node. In terms of expressiveness, we show that the Transformer is capable of executing path-finding by embedding the adjacency and reachability matrices within its weights. Our theoretical analysis of the gradient-based learning dynamic of the Transformer reveals that the Transformer is capable of learning both the adjacency matrix and a limited form of the reachability matrix. These theoretical insights are then validated through experiments, which demonstrate that the Transformer indeed learns the adjacency matrix and an incomplete reachability matrix, which aligns with the predictions made in our theoretical analysis. Additionally, when applying our methodology to a real-world planning benchmark, called Blocksworld, our observations remain consistent. Our theoretical and empirical analyses further unveil a potential limitation of Transformer in path-finding: it cannot identify reachability relationships through transitivity, and thus would fail when path concatenation is needed to generate a path. In summary, our findings shed new light on how the internal mechanisms of autoregressive learning enable planning in networks. This study may contribute to our understanding of the general planning capabilities in other related domains.