Yemin Shi

Ge Zhang, Yemin Shi, Ruibo Liu et al. · deepmind

Instruction tuning is widely recognized as a key technique for building generalist language models, which has attracted the attention of researchers and the public with the release of InstructGPT~\citep{ouyang2022training} and ChatGPT\footnote{\url{https://chat.openai.com/}}. Despite impressive progress in English-oriented large-scale language models (LLMs), it is still under-explored whether English-based foundation LLMs can perform similarly on multilingual tasks compared to English tasks with well-designed instruction tuning and how we can construct the corpora needed for the tuning. To remedy this gap, we propose the project as an attempt to create a Chinese instruction dataset by various methods adapted to the intrinsic characteristics of 4 sub-tasks. We collect around 200k Chinese instruction tuning samples, which have been manually checked to guarantee high quality. We also summarize the existing English and Chinese instruction corpora and briefly describe some potential applications of the newly constructed Chinese instruction corpora. The resulting \textbf{C}hinese \textbf{O}pen \textbf{I}nstruction \textbf{G}eneralist (\textbf{COIG}) corpora are available in Huggingface\footnote{\url{https://huggingface.co/datasets/BAAI/COIG}} and Github\footnote{\url{https://github.com/BAAI-Zlab/COIG}}, and will be continuously updated.

Guangyao Chen, Siwei Dong, Yu Shu et al.

Large language models (LLMs) have enabled remarkable advances in automated task-solving with multi-agent systems. However, most existing LLM-based multi-agent approaches rely on predefined agents to handle simple tasks, limiting the adaptability of multi-agent collaboration to different scenarios. Therefore, we introduce AutoAgents, an innovative framework that adaptively generates and coordinates multiple specialized agents to build an AI team according to different tasks. Specifically, AutoAgents couples the relationship between tasks and roles by dynamically generating multiple required agents based on task content and planning solutions for the current task based on the generated expert agents. Multiple specialized agents collaborate with each other to efficiently accomplish tasks. Concurrently, an observer role is incorporated into the framework to reflect on the designated plans and agents' responses and improve upon them. Our experiments on various benchmarks demonstrate that AutoAgents generates more coherent and accurate solutions than the existing multi-agent methods. This underscores the significance of assigning different roles to different tasks and of team cooperation, offering new perspectives for tackling complex tasks. The repository of this project is available at https://github.com/Link-AGI/AutoAgents.

Yu Shu, Siwei Dong, Guangyao Chen et al.

Multi-modal large language models have garnered significant interest recently. Though, most of the works focus on vision-language multi-modal models providing strong capabilities in following vision-and-language instructions. However, we claim that speech is also an important modality through which humans interact with the world. Hence, it is crucial for a general-purpose assistant to be able to follow multi-modal speech-and-language instructions. In this work, we propose Large Language and Speech Model (LLaSM). LLaSM is an end-to-end trained large multi-modal speech-language model with cross-modal conversational abilities, capable of following speech-and-language instructions. Our early experiments show that LLaSM demonstrates a more convenient and natural way for humans to interact with artificial intelligence. Specifically, we also release a large Speech Instruction Following dataset LLaSM-Audio-Instructions. Code and demo are available at https://github.com/LinkSoul-AI/LLaSM and https://huggingface.co/spaces/LinkSoul/LLaSM. The LLaSM-Audio-Instructions dataset is available at https://huggingface.co/datasets/LinkSoul/LLaSM-Audio-Instructions.

Jingkun An, Yinghao Zhu, Zongjian Li et al.

Text-to-Image (T2I) diffusion models have achieved remarkable success in image generation. Despite their progress, challenges remain in both prompt-following ability, image quality and lack of high-quality datasets, which are essential for refining these models. As acquiring labeled data is costly, we introduce AGFSync, a framework that enhances T2I diffusion models through Direct Preference Optimization (DPO) in a fully AI-driven approach. AGFSync utilizes Vision-Language Models (VLM) to assess image quality across style, coherence, and aesthetics, generating feedback data within an AI-driven loop. By applying AGFSync to leading T2I models such as SD v1.4, v1.5, and SDXL-base, our extensive experiments on the TIFA dataset demonstrate notable improvements in VQA scores, aesthetic evaluations, and performance on the HPSv2 benchmark, consistently outperforming the base models. AGFSync's method of refining T2I diffusion models paves the way for scalable alignment techniques. Our code and dataset are publicly available at https://anjingkun.github.io/AGFSync.

Yemin Shi, Yu Shu, Siwei Dong et al.

A voice AI agent that blends seamlessly into daily life would interact with humans in an autonomous, real-time, and emotionally expressive manner. Rather than merely reacting to commands, it would continuously listen, reason, and respond proactively, fostering fluid, dynamic, and emotionally resonant interactions. We introduce Voila, a family of large voice-language foundation models that make a step towards this vision. Voila moves beyond traditional pipeline systems by adopting a new end-to-end architecture that enables full-duplex, low-latency conversations while preserving rich vocal nuances such as tone, rhythm, and emotion. It achieves a response latency of just 195 milliseconds, surpassing the average human response time. Its hierarchical multi-scale Transformer integrates the reasoning capabilities of large language models (LLMs) with powerful acoustic modeling, enabling natural, persona-aware voice generation -- where users can simply write text instructions to define the speaker's identity, tone, and other characteristics. Moreover, Voila supports over one million pre-built voices and efficient customization of new ones from brief audio samples as short as 10 seconds. Beyond spoken dialogue, Voila is designed as a unified model for a wide range of voice-based applications, including automatic speech recognition (ASR), Text-to-Speech (TTS), and, with minimal adaptation, multilingual speech translation. Voila is fully open-sourced to support open research and accelerate progress toward next-generation human-machine interactions.

Yemin Shi, Yonghong Tian, Yaowei Wang et al.

Despite a lot of research efforts devoted in recent years, how to efficiently learn long-term dependencies from sequences still remains a pretty challenging task. As one of the key models for sequence learning, recurrent neural network (RNN) and its variants such as long short term memory (LSTM) and gated recurrent unit (GRU) are still not powerful enough in practice. One possible reason is that they have only feedforward connections, which is different from the biological neural system that is typically composed of both feedforward and feedback connections. To address this problem, this paper proposes a biologically-inspired deep network, called shuttleNet\footnote{Our code is available at \url{https://github.com/shiyemin/shuttlenet}}. Technologically, the shuttleNet consists of several processors, each of which is a GRU while associated with multiple groups of cells and states. Unlike traditional RNNs, all processors inside shuttleNet are loop connected to mimic the brain's feedforward and feedback connections, in which they are shared across multiple pathways in the loop connection. Attention mechanism is then employed to select the best information flow pathway. Extensive experiments conducted on two benchmark datasets (i.e UCF101 and HMDB51) show that we can beat state-of-the-art methods by simply embedding shuttleNet into a CNN-RNN framework.

Yi Gui, Zhen Li, Yao Wan et al.

Automatically generating webpage code from webpage designs can significantly reduce the workload of front-end developers, and recent Multimodal Large Language Models (MLLMs) have shown promising potential in this area. However, our investigation reveals that most existing MLLMs are constrained by the absence of high-quality, large-scale, real-world datasets, resulting in inadequate performance in automated webpage code generation. To fill this gap, this paper introduces WebCode2M, a new dataset comprising 2.56 million instances, each containing a design image along with the corresponding webpage code and layout details. Sourced from real-world web resources, WebCode2M offers a rich and valuable dataset for webpage code generation across a variety of applications. The dataset quality is ensured by a scoring model that filters out instances with aesthetic deficiencies or other incomplete elements. To validate the effectiveness of WebCode2M, we introduce a baseline model based on the Vision Transformer (ViT), named WebCoder, and establish a benchmark for fair comparison. Additionally, we introduce a new metric, TreeBLEU, to measure the structural hierarchy recall. The benchmarking results demonstrate that our dataset significantly improves the ability of MLLMs to generate code from webpage designs, confirming its effectiveness and usability for future applications in front-end design tools. Finally, we highlight several practical challenges introduced by our dataset, calling for further research. The code and dataset are publicly available at our project homepage: https://webcode2m.github.io.

Jiannan Xiang, Guangyi Liu, Yi Gu et al.

World models simulate future states of the world in response to different actions. They facilitate interactive content creation and provides a foundation for grounded, long-horizon reasoning. Current foundation models do not fully meet the capabilities of general world models: large language models (LLMs) are constrained by their reliance on language modality and their limited understanding of the physical world, while video models lack interactive action control over the world simulations. This paper makes a step towards building a general world model by introducing Pandora, a hybrid autoregressive-diffusion model that simulates world states by generating videos and allows real-time control with free-text actions. Pandora achieves domain generality, video consistency, and controllability through large-scale pretraining and instruction tuning. Crucially, Pandora bypasses the cost of training-from-scratch by integrating a pretrained LLM (7B) and a pretrained video model, requiring only additional lightweight finetuning. We illustrate extensive outputs by Pandora across diverse domains (indoor/outdoor, natural/urban, human/robot, 2D/3D, etc.). The results indicate great potential of building stronger general world models with larger-scale training.

Yizhi Li, Ruibin Yuan, Ge Zhang et al.

Self-supervised learning (SSL) has recently emerged as a promising paradigm for training generalisable models on large-scale data in the fields of vision, text, and speech. Although SSL has been proven effective in speech and audio, its application to music audio has yet to be thoroughly explored. This is partially due to the distinctive challenges associated with modelling musical knowledge, particularly tonal and pitched characteristics of music. To address this research gap, we propose an acoustic Music undERstanding model with large-scale self-supervised Training (MERT), which incorporates teacher models to provide pseudo labels in the masked language modelling (MLM) style acoustic pre-training. In our exploration, we identified an effective combination of teacher models, which outperforms conventional speech and audio approaches in terms of performance. This combination includes an acoustic teacher based on Residual Vector Quantisation - Variational AutoEncoder (RVQ-VAE) and a musical teacher based on the Constant-Q Transform (CQT). Furthermore, we explore a wide range of settings to overcome the instability in acoustic language model pre-training, which allows our designed paradigm to scale from 95M to 330M parameters. Experimental results indicate that our model can generalise and perform well on 14 music understanding tasks and attain state-of-the-art (SOTA) overall scores.

Gang Yu, Zhongzhi Yu, Yemin Shi et al.

Respiratory diseases, including asthma, bronchitis, pneumonia, and upper respiratory tract infection (RTI), are among the most common diseases in clinics. The similarities among the symptoms of these diseases precludes prompt diagnosis upon the patients' arrival. In pediatrics, the patients' limited ability in expressing their situation makes precise diagnosis even harder. This becomes worse in primary hospitals, where the lack of medical imaging devices and the doctors' limited experience further increase the difficulty of distinguishing among similar diseases. In this paper, a pediatric fine-grained diagnosis-assistant system is proposed to provide prompt and precise diagnosis using solely clinical notes upon admission, which would assist clinicians without changing the diagnostic process. The proposed system consists of two stages: a test result structuralization stage and a disease identification stage. The first stage structuralizes test results by extracting relevant numerical values from clinical notes, and the disease identification stage provides a diagnosis based on text-form clinical notes and the structured data obtained from the first stage. A novel deep learning algorithm was developed for the disease identification stage, where techniques including adaptive feature infusion and multi-modal attentive fusion were introduced to fuse structured and text data together. Clinical notes from over 12000 patients with respiratory diseases were used to train a deep learning model, and clinical notes from a non-overlapping set of about 1800 patients were used to evaluate the performance of the trained model. The average precisions (AP) for pneumonia, RTI, bronchitis and asthma are 0.878, 0.857, 0.714, and 0.825, respectively, achieving a mean AP (mAP) of 0.819.

Jichang Li, Guanbin Li, Yemin Shi et al.

In semi-supervised domain adaptation, a few labeled samples per class in the target domain guide features of the remaining target samples to aggregate around them. However, the trained model cannot produce a highly discriminative feature representation for the target domain because the training data is dominated by labeled samples from the source domain. This could lead to disconnection between the labeled and unlabeled target samples as well as misalignment between unlabeled target samples and the source domain. In this paper, we propose a novel approach called Cross-domain Adaptive Clustering to address this problem. To achieve both inter-domain and intra-domain adaptation, we first introduce an adversarial adaptive clustering loss to group features of unlabeled target data into clusters and perform cluster-wise feature alignment across the source and target domains. We further apply pseudo labeling to unlabeled samples in the target domain and retain pseudo-labels with high confidence. Pseudo labeling expands the number of ``labeled" samples in each class in the target domain, and thus produces a more robust and powerful cluster core for each class to facilitate adversarial learning. Extensive experiments on benchmark datasets, including DomainNet, Office-Home and Office, demonstrate that our proposed approach achieves the state-of-the-art performance in semi-supervised domain adaptation.

Guangyao Chen, Limeng Qiao, Yemin Shi et al.

Open set recognition is an emerging research area that aims to simultaneously classify samples from predefined classes and identify the rest as 'unknown'. In this process, one of the key challenges is to reduce the risk of generalizing the inherent characteristics of numerous unknown samples learned from a small amount of known data. In this paper, we propose a new concept, Reciprocal Point, which is the potential representation of the extra-class space corresponding to each known category. The sample can be classified to known or unknown by the otherness with reciprocal points. To tackle the open set problem, we offer a novel open space risk regularization term. Based on the bounded space constructed by reciprocal points, the risk of unknown is reduced through multi-category interaction. The novel learning framework called Reciprocal Point Learning (RPL), which can indirectly introduce the unknown information into the learner with only known classes, so as to learn more compact and discriminative representations. Moreover, we further construct a new large-scale challenging aircraft dataset for open set recognition: Aircraft 300 (Air-300). Extensive experiments on multiple benchmark datasets indicate that our framework is significantly superior to other existing approaches and achieves state-of-the-art performance on standard open set benchmarks.

Zhongzhi Yu, Yemin Shi, Tiejun Huang et al.

This paper presents a novel network compression framework Kernel Quantization (KQ), targeting to efficiently convert any pre-trained full-precision convolutional neural network (CNN) model into a low-precision version without significant performance loss. Unlike existing methods struggling with weight bit-length, KQ has the potential in improving the compression ratio by considering the convolution kernel as the quantization unit. Inspired by the evolution from weight pruning to filter pruning, we propose to quantize in both kernel and weight level. Instead of representing each weight parameter with a low-bit index, we learn a kernel codebook and replace all kernels in the convolution layer with corresponding low-bit indexes. Thus, KQ can represent the weight tensor in the convolution layer with low-bit indexes and a kernel codebook with limited size, which enables KQ to achieve significant compression ratio. Then, we conduct a 6-bit parameter quantization on the kernel codebook to further reduce redundancy. Extensive experiments on the ImageNet classification task prove that KQ needs 1.05 and 1.62 bits on average in VGG and ResNet18, respectively, to represent each parameter in the convolution layer and achieves the state-of-the-art compression ratio with little accuracy loss.

Limeng Qiao, Yemin Shi, Jia Li et al.

Few-shot learning, which aims at extracting new concepts rapidly from extremely few examples of novel classes, has been featured into the meta-learning paradigm recently. Yet, the key challenge of how to learn a generalizable classifier with the capability of adapting to specific tasks with severely limited data still remains in this domain. To this end, we propose a Transductive Episodic-wise Adaptive Metric (TEAM) framework for few-shot learning, by integrating the meta-learning paradigm with both deep metric learning and transductive inference. With exploring the pairwise constraints and regularization prior within each task, we explicitly formulate the adaptation procedure into a standard semi-definite programming problem. By solving the problem with its closed-form solution on the fly with the setup of transduction, our approach efficiently tailors an episodic-wise metric for each task to adapt all features from a shared task-agnostic embedding space into a more discriminative task-specific metric space. Moreover, we further leverage an attention-based bi-directional similarity strategy for extracting the more robust relationship between queries and prototypes. Extensive experiments on three benchmark datasets show that our framework is superior to other existing approaches and achieves the state-of-the-art performance in the few-shot literature.

Yu Shu, Yemin Shi, Yaowei Wang et al.

Most of the existing recognition algorithms are proposed for closed set scenarios, where all categories are known beforehand. However, in practice, recognition is essentially an open set problem. There are categories we know called "knowns", and there are more we do not know called "unknowns". Enumerating all categories beforehand is never possible, consequently it is infeasible to prepare sufficient training samples for those unknowns. Applying closed set recognition methods will naturally lead to unseen-category errors. To address this problem, we propose the prototype based Open Deep Network (P-ODN) for open set recognition tasks. Specifically, we introduce prototype learning into open set recognition. Prototypes and prototype radiuses are trained jointly to guide a CNN network to derive more discriminative features. Then P-ODN detects the unknowns by applying a multi-class triplet thresholding method based on the distance metric between features and prototypes. Manual labeling the unknowns which are detected in the previous process as new categories. Predictors for new categories are added to the classification layer to "open" the deep neural networks to incorporate new categories dynamically. The weights of new predictors are initialized exquisitely by applying a distances based algorithm to transfer the learned knowledge. Consequently, this initialization method speed up the fine-tuning process and reduce the samples needed to train new predictors. Extensive experiments show that P-ODN can effectively detect unknowns and needs only few samples with human intervention to recognize a new category. In the real world scenarios, our method achieves state-of-the-art performance on the UCF11, UCF50, UCF101 and HMDB51 datasets.

Yu Shu, Yemin Shi, Yaowei Wang et al.

In recent years, the performance of action recognition has been significantly improved with the help of deep neural networks. Most of the existing action recognition works hold the \textit{closed-set} assumption that all action categories are known beforehand while deep networks can be well trained for these categories. However, action recognition in the real world is essentially an \textit{open-set} problem, namely, it is impossible to know all action categories beforehand and consequently infeasible to prepare sufficient training samples for those emerging categories. In this case, applying closed-set recognition methods will definitely lead to unseen-category errors. To address this challenge, we propose the Open Deep Network (ODN) for the open-set action recognition task. Technologically, ODN detects new categories by applying a multi-class triplet thresholding method, and then dynamically reconstructs the classification layer and "opens" the deep network by adding predictors for new categories continually. In order to transfer the learned knowledge to the new category, two novel methods, Emphasis Initialization and Allometry Training, are adopted to initialize and incrementally train the new predictor so that only few samples are needed to fine-tune the model. Extensive experiments show that ODN can effectively detect and recognize new categories with little human intervention, thus applicable to the open-set action recognition tasks in the real world. Moreover, ODN can even achieve comparable performance to some closed-set methods.

Yemin Shi, Yonghong Tian, Yaowei Wang et al.

By extracting spatial and temporal characteristics in one network, the two-stream ConvNets can achieve the state-of-the-art performance in action recognition. However, such a framework typically suffers from the separately processing of spatial and temporal information between the two standalone streams and is hard to capture long-term temporal dependence of an action. More importantly, it is incapable of finding the salient portions of an action, say, the frames that are the most discriminative to identify the action. To address these problems, a \textbf{j}oint \textbf{n}etwork based \textbf{a}ttention (JNA) is proposed in this study. We find that the fully-connected fusion, branch selection and spatial attention mechanism are totally infeasible for action recognition. Thus in our joint network, the spatial and temporal branches share some information during the training stage. We also introduce an attention mechanism on the temporal domain to capture the long-term dependence meanwhile finding the salient portions. Extensive experiments are conducted on two benchmark datasets, UCF101 and HMDB51. Experimental results show that our method can improve the action recognition performance significantly and achieves the state-of-the-art results on both datasets.

Yemin Shi, Yonghong Tian, Yaowei Wang et al.

Learning the spatial-temporal representation of motion information is crucial to human action recognition. Nevertheless, most of the existing features or descriptors cannot capture motion information effectively, especially for long-term motion. To address this problem, this paper proposes a long-term motion descriptor called sequential Deep Trajectory Descriptor (sDTD). Specifically, we project dense trajectories into two-dimensional planes, and subsequently a CNN-RNN network is employed to learn an effective representation for long-term motion. Unlike the popular two-stream ConvNets, the sDTD stream is introduced into a three-stream framework so as to identify actions from a video sequence. Consequently, this three-stream framework can simultaneously capture static spatial features, short-term motion and long-term motion in the video. Extensive experiments were conducted on three challenging datasets: KTH, HMDB51 and UCF101. Experimental results show that our method achieves state-of-the-art performance on the KTH and UCF101 datasets, and is comparable to the state-of-the-art methods on the HMDB51 dataset.