Shuo Wang

Chaojun Xiao, Zhengyan Zhang, Chenyang Song et al. · tencent-ai, tsinghua

Advancements in LLMs have recently unveiled challenges tied to computational efficiency and continual scalability due to their requirements of huge parameters, making the applications and evolution of these models on devices with limited computation resources and scenarios requiring various abilities increasingly cumbersome. Inspired by modularity within the human brain, there is a growing tendency to decompose LLMs into numerous functional modules, allowing for inference with part of modules and dynamic assembly of modules to tackle complex tasks, such as mixture-of-experts. To highlight the inherent efficiency and composability of the modular approach, we coin the term brick to represent each functional module, designating the modularized structure as configurable foundation models. In this paper, we offer a comprehensive overview and investigation of the construction, utilization, and limitation of configurable foundation models. We first formalize modules into emergent bricks - functional neuron partitions that emerge during the pre-training phase, and customized bricks - bricks constructed via additional post-training to improve the capabilities and knowledge of LLMs. Based on diverse functional bricks, we further present four brick-oriented operations: retrieval and routing, merging, updating, and growing. These operations allow for dynamic configuration of LLMs based on instructions to handle complex tasks. To verify our perspective, we conduct an empirical analysis on widely-used LLMs. We find that the FFN layers follow modular patterns with functional specialization of neurons and functional neuron partitions. Finally, we highlight several open issues and directions for future research. Overall, this paper aims to offer a fresh modular perspective on existing LLM research and inspire the future creation of more efficient and scalable foundational models.

Shuo Wang, Jing Li, Zibo Zhao et al.

Holistic scene understanding includes semantic segmentation, surface normal estimation, object boundary detection, depth estimation, etc. The key aspect of this problem is to learn representation effectively, as each subtask builds upon not only correlated but also distinct attributes. Inspired by visual-prompt tuning, we propose a Task-Specific Prompts Transformer, dubbed TSP-Transformer, for holistic scene understanding. It features a vanilla transformer in the early stage and tasks-specific prompts transformer encoder in the lateral stage, where tasks-specific prompts are augmented. By doing so, the transformer layer learns the generic information from the shared parts and is endowed with task-specific capacity. First, the tasks-specific prompts serve as induced priors for each task effectively. Moreover, the task-specific prompts can be seen as switches to favor task-specific representation learning for different tasks. Extensive experiments on NYUD-v2 and PASCAL-Context show that our method achieves state-of-the-art performance, validating the effectiveness of our method for holistic scene understanding. We also provide our code in the following link https://github.com/tb2-sy/TSP-Transformer.

Wang Boya, Wang Shuo, Ye Dong et al.

With the frequent use of self-supervised monocular depth estimation in robotics and autonomous driving, the model's efficiency is becoming increasingly important. Most current approaches apply much larger and more complex networks to improve the precision of depth estimation. Some researchers incorporated Transformer into self-supervised monocular depth estimation to achieve better performance. However, this method leads to high parameters and high computation. We present a fully convolutional depth estimation network using contextual feature fusion. Compared to UNet++ and HRNet, we use high-resolution and low-resolution features to reserve information on small targets and fast-moving objects instead of long-range fusion. We further promote depth estimation results employing lightweight channel attention based on convolution in the decoder stage. Our method reduces the parameters without sacrificing accuracy. Experiments on the KITTI benchmark show that our method can get better results than many large models, such as Monodepth2, with only 30 parameters. The source code is available at https://github.com/boyagesmile/DNA-Depth.

Xiaoqin Zhang, Ziwei Huang, Jingjing Zheng et al.

The task of grasp pattern recognition aims to derive the applicable grasp types of an object according to the visual information. Current state-of-the-art methods ignore category information of objects which is crucial for grasp pattern recognition. This paper presents a novel dual-branch convolutional neural network (DcnnGrasp) to achieve joint learning of object category classification and grasp pattern recognition. DcnnGrasp takes object category classification as an auxiliary task to improve the effectiveness of grasp pattern recognition. Meanwhile, a new loss function called joint cross-entropy with an adaptive regularizer is derived through maximizing a posterior, which significantly improves the model performance. Besides, based on the new loss function, a training strategy is proposed to maximize the collaborative learning of the two tasks. The experiment was performed on five household objects datasets including the RGB-D Object dataset, Hit-GPRec dataset, Amsterdam library of object images (ALOI), Columbia University Image Library (COIL-100), and MeganePro dataset 1. The experimental results demonstrated that the proposed method can achieve competitive performance on grasp pattern recognition with several state-of-the-art methods. Specifically, our method even outperformed the second-best one by nearly 15% in terms of global accuracy for the case of testing a novel object on the RGB-D Object dataset.

Jinda Lu, Shuo Wang, Yanbin Hao et al.

Recent adaptations can boost the low-shot capability of Contrastive Vision-Language Pre-training (CLIP) by effectively facilitating knowledge transfer. However, these adaptation methods are usually operated on the global view of an input image, and thus biased perception of partial local details of the image. To solve this problem, we propose a Visual Content Refinement (VCR) before the adaptation calculation during the test stage. Specifically, we first decompose the test image into different scales to shift the feature extractor's attention to the details of the image. Then, we select the image view with the max prediction margin in each scale to filter out the noisy image views, where the prediction margins are calculated from the pre-trained CLIP model. Finally, we merge the content of the aforementioned selected image views based on their scales to construct a new robust representation. Thus, the merged content can be directly used to help the adapter focus on both global and local parts without any extra training parameters. We apply our method to 3 popular low-shot benchmark tasks with 13 datasets and achieve a significant improvement over state-of-the-art methods. For example, compared to the baseline (Tip-Adapter) on the few-shot classification task, our method achieves about 2\% average improvement for both training-free and training-need settings.

Zheng Tang, Milind Naphade, Stan Birchfield et al.

In comparison with person re-identification (ReID), which has been widely studied in the research community, vehicle ReID has received less attention. Vehicle ReID is challenging due to 1) high intra-class variability (caused by the dependency of shape and appearance on viewpoint), and 2) small inter-class variability (caused by the similarity in shape and appearance between vehicles produced by different manufacturers). To address these challenges, we propose a Pose-Aware Multi-Task Re-Identification (PAMTRI) framework. This approach includes two innovations compared with previous methods. First, it overcomes viewpoint-dependency by explicitly reasoning about vehicle pose and shape via keypoints, heatmaps and segments from pose estimation. Second, it jointly classifies semantic vehicle attributes (colors and types) while performing ReID, through multi-task learning with the embedded pose representations. Since manually labeling images with detailed pose and attribute information is prohibitive, we create a large-scale highly randomized synthetic dataset with automatically annotated vehicle attributes for training. Extensive experiments validate the effectiveness of each proposed component, showing that PAMTRI achieves significant improvement over state-of-the-art on two mainstream vehicle ReID benchmarks: VeRi and CityFlow-ReID. Code and models are available at https://github.com/NVlabs/PAMTRI.

Peng Hao, Chaofan Zhang, Dingzhe Li et al.

Significant progress has been made in vision-language models. However, language-conditioned robotic manipulation for contact-rich tasks remains underexplored, particularly in terms of tactile sensing. To address this gap, we introduce the Tactile-Language-Action (TLA) model, which effectively processes sequential tactile feedback via cross-modal language grounding to enable robust policy generation in contact-intensive scenarios. In addition, we construct a comprehensive dataset that contains 24k pairs of tactile action instruction data, customized for fingertip peg-in-hole assembly, providing essential resources for TLA training and evaluation. Our results show that TLA significantly outperforms traditional imitation learning methods (e.g., diffusion policy) in terms of effective action generation and action accuracy, while demonstrating strong generalization capabilities by achieving over 85\% success rate on previously unseen assembly clearances and peg shapes. We publicly release all data and code in the hope of advancing research in language-conditioned tactile manipulation skill learning. Project website: https://sites.google.com/view/tactile-language-action/

Xingyu Zhu, Shuo Wang, Jinda Lu et al.

Few-shot learning (FSL) based on manifold regularization aims to improve the recognition capacity of novel objects with limited training samples by mixing two samples from different categories with a blending factor. However, this mixing operation weakens the feature representation due to the linear interpolation and the overlooking of the importance of specific channels. To solve these issues, this paper proposes attentive feature regularization (AFR) which aims to improve the feature representativeness and discriminability. In our approach, we first calculate the relations between different categories of semantic labels to pick out the related features used for regularization. Then, we design two attention-based calculations at both the instance and channel levels. These calculations enable the regularization procedure to focus on two crucial aspects: the feature complementarity through adaptive interpolation in related categories and the emphasis on specific feature channels. Finally, we combine these regularization strategies to significantly improve the classifier performance. Empirical studies on several popular FSL benchmarks demonstrate the effectiveness of AFR, which improves the recognition accuracy of novel categories without the need to retrain any feature extractor, especially in the 1-shot setting. Furthermore, the proposed AFR can seamlessly integrate into other FSL methods to improve classification performance.

Chihang Wang, Yuxin Dong, Zhenhong Zhang et al.

This paper focuses on the development of an advanced intelligent article scoring system that not only assesses the overall quality of written work but also offers detailed feature-based scoring tailored to various article genres. By integrating the pre-trained BERT model with the large language model Chat-GPT, the system gains a deep understanding of both the content and structure of the text, enabling it to provide a thorough evaluation along with targeted suggestions for improvement. Experimental results demonstrate that this system outperforms traditional scoring methods across multiple public datasets, particularly in feature-based assessments, offering a more accurate reflection of the quality of different article types. Moreover, the system generates personalized feedback to assist users in enhancing their writing skills, underscoring the potential and practical value of automated scoring technologies in educational contexts.

Chuanbo Hu, Jacob Thrasher, Wenqi Li et al.

Diagnosing autism spectrum disorder (ASD) by identifying abnormal speech patterns from examiner-patient dialogues presents significant challenges due to the subtle and diverse manifestations of speech-related symptoms in affected individuals. This study presents a comprehensive approach to identify distinctive speech patterns through the analysis of examiner-patient dialogues. Utilizing a dataset of recorded dialogues, we extracted 40 speech-related features, categorized into frequency, zero-crossing rate, energy, spectral characteristics, Mel Frequency Cepstral Coefficients (MFCCs), and balance. These features encompass various aspects of speech such as intonation, volume, rhythm, and speech rate, reflecting the complex nature of communicative behaviors in ASD. We employed machine learning for both classification and regression tasks to analyze these speech features. The classification model aimed to differentiate between ASD and non-ASD cases, achieving an accuracy of 87.75%. Regression models were developed to predict speech pattern related variables and a composite score from all variables, facilitating a deeper understanding of the speech dynamics associated with ASD. The effectiveness of machine learning in interpreting intricate speech patterns and the high classification accuracy underscore the potential of computational methods in supporting the diagnostic processes for ASD. This approach not only aids in early detection but also contributes to personalized treatment planning by providing insights into the speech and communication profiles of individuals with ASD.

Chuanbo Hu, Wenqi Li, Mindi Ruan et al.

Diagnosing language disorders associated with autism is a complex challenge, often hampered by the subjective nature and variability of traditional assessment methods. Traditional diagnostic methods not only require intensive human effort but also often result in delayed interventions due to their lack of speed and precision. In this study, we explored the application of ChatGPT, a large language model, to overcome these obstacles by enhancing sensitivity and profiling linguistic features for autism diagnosis. This research utilizes ChatGPT natural language processing capabilities to simplify and improve the diagnostic process, focusing on identifying autism related language patterns. Specifically, we compared ChatGPT performance with that of conventional supervised learning models, including BERT, a model acclaimed for its effectiveness in various natural language processing tasks. We showed that ChatGPT substantially outperformed these models, achieving over 10% improvement in both sensitivity and positive predictive value, in a zero shot learning configuration. The findings underscore the model potential as a diagnostic tool, combining accuracy and applicability. We identified ten key features of autism associated language disorders across scenarios. Features such as echolalia, pronoun reversal, and atypical language usage play a critical role in diagnosing ASD and informing tailored treatment plans. Together, our findings advocate for adopting sophisticated AI tools like ChatGPT in clinical settings to assess and diagnose developmental disorders. Our approach promises enhanced diagnostic precision and supports personalized medicine, potentially transforming the evaluation landscape for autism and similar neurological conditions.

Xujia Wang, Haiyan Zhao, Shuo Wang et al.

Parameter-Efficient Fine-Tuning (PEFT) methods like LoRA have significantly improved the adaptation of LLMs to downstream tasks in a resource-efficient manner. However, in multi-task scenarios, challenges such as training imbalance and the seesaw effect frequently emerge. Mixture-of-LoRA (MoLoRA), which combines LoRA with sparse Mixture-of-Experts, mitigates some of these issues by promoting task-specific learning across experts. Despite this, MoLoRA remains inefficient in terms of training speed, parameter utilization, and overall multi-task performance. In this paper, we propose Mixture of Asymmetric Low-Rank Adaptaion (MALoRA), a flexible fine-tuning framework that leverages asymmetric optimization across LoRA experts. MALoRA reduces the number of trainable parameters by 30% to 48%, increases training speed by 1.2x, and matches the computational efficiency of single-task LoRA models. Additionally, MALoRA addresses overfitting issues commonly seen in high-rank configurations, enhancing performance stability. Extensive experiments across diverse multi-task learning scenarios demonstrate that MALoRA consistently outperforms all baseline methods in both inter-domain and intra-domain tasks.

Xiaobo Hu, Youfang Lin, Yue Liu et al.

Visual reinforcement learning has proven effective in solving control tasks with high-dimensional observations. However, extracting reliable and generalizable representations from vision-based observations remains a central challenge. Inspired by the human thought process, when the representation extracted from the observation can predict the future and trace history, the representation is reliable and accurate in comprehending the environment. Based on this concept, we introduce a Bidirectional Transition (BiT) model, which leverages the ability to bidirectionally predict environmental transitions both forward and backward to extract reliable representations. Our model demonstrates competitive generalization performance and sample efficiency on two settings of the DeepMind Control suite. Additionally, we utilize robotic manipulation and CARLA simulators to demonstrate the wide applicability of our method.

Hainan Li, Renshuai Tao, Jun Li et al.

Few-shot learning is an interesting and challenging study, which enables machines to learn from few samples like humans. Existing studies rarely exploit auxiliary information from large amount of unlabeled data. Self-supervised learning is emerged as an efficient method to utilize unlabeled data. Existing self-supervised learning methods always rely on the combination of geometric transformations for the single sample by augmentation, while seriously neglect the endogenous correlation information among different samples that is the same important for the task. In this work, we propose a Graph-driven Clustering (GC), a novel augmentation-free method for self-supervised learning, which does not rely on any auxiliary sample and utilizes the endogenous correlation information among input samples. Besides, we propose Multi-pretext Attention Network (MAN), which exploits a specific attention mechanism to combine the traditional augmentation-relied methods and our GC, adaptively learning their optimized weights to improve the performance and enabling the feature extractor to obtain more universal representations. We evaluate our MAN extensively on miniImageNet and tieredImageNet datasets and the results demonstrate that the proposed method outperforms the state-of-the-art (SOTA) relevant methods.

Ye Bi, Shuo Wang, Zhongrui Fan

The cataloging of product listings is a fundamental problem for most e-commerce platforms. Despite promising results obtained by unimodal-based methods, it can be expected that their performance can be further boosted by the consideration of multimodal product information. In this study, we investigated a multimodal late fusion approach based on text and image modalities to categorize e-commerce products on Rakuten. Specifically, we developed modal specific state-of-the-art deep neural networks for each input modal, and then fused them at the decision level. Experimental results on Multimodal Product Classification Task of SIGIR 2020 E-Commerce Workshop Data Challenge demonstrate the superiority and effectiveness of our proposed method compared with unimodal and other multimodal methods. Our team named pa_curis won the 1st place with a macro-F1 of 0.9144 on the final leaderboard.

Ye Bi, Shuo Wang, Zhongrui Fan

The animated Graphical Interchange Format (GIF) images have been widely used on social media as an intuitive way of expression emotion. Given their expressiveness, GIFs offer a more nuanced and precise way to convey emotions. In this paper, we present our solution for the EmotionGIF 2020 challenge, the shared task of SocialNLP 2020. To recommend GIF categories for unlabeled tweets, we regarded this problem as a kind of matching tasks and proposed a learning to rank framework based on Bidirectional Encoder Representations from Transformer (BERT) and LightGBM. Our team won the 4th place with a Mean Average Precision @ 6 (MAP@6) score of 0.5394 on the round 1 leaderboard.

Shaowei Cui, Rui Wang, Junhang Wei et al.

Humans can quickly determine the force required to grasp a deformable object to prevent its sliding or excessive deformation through vision and touch, which is still a challenging task for robots. To address this issue, we propose a novel 3D convolution-based visual-tactile fusion deep neural network (C3D-VTFN) to evaluate the grasp state of various deformable objects in this paper. Specifically, we divide the grasp states of deformable objects into three categories of sliding, appropriate and excessive. Also, a dataset for training and testing the proposed network is built by extensive grasping and lifting experiments with different widths and forces on 16 various deformable objects with a robotic arm equipped with a wrist camera and a tactile sensor. As a result, a classification accuracy as high as 99.97% is achieved. Furthermore, some delicate grasp experiments based on the proposed network are implemented in this paper. The experimental results demonstrate that the C3D-VTFN is accurate and efficient enough for grasp state assessment, which can be widely applied to automatic force control, adaptive grasping, and other visual-tactile spatiotemporal sequence learning problems.

Jiamiao Xu, Fangzhao Wang, Qinmu Peng et al.

Low-rank Multi-view Subspace Learning (LMvSL) has shown great potential in cross-view classification in recent years. Despite their empirical success, existing LMvSL based methods are incapable of well handling view discrepancy and discriminancy simultaneously, which thus leads to the performance degradation when there is a large discrepancy among multi-view data. To circumvent this drawback, motivated by the block-diagonal representation learning, we propose Structured Low-rank Matrix Recovery (SLMR), a unique method of effectively removing view discrepancy and improving discriminancy through the recovery of structured low-rank matrix. Furthermore, recent low-rank modeling provides a satisfactory solution to address data contaminated by predefined assumptions of noise distribution, such as Gaussian or Laplacian distribution. However, these models are not practical since complicated noise in practice may violate those assumptions and the distribution is generally unknown in advance. To alleviate such limitation, modal regression is elegantly incorporated into the framework of SLMR (term it MR-SLMR). Different from previous LMvSL based methods, our MR-SLMR can handle any zero-mode noise variable that contains a wide range of noise, such as Gaussian noise, random noise and outliers. The alternating direction method of multipliers (ADMM) framework and half-quadratic theory are used to efficiently optimize MR-SLMR. Experimental results on four public databases demonstrate the superiority of MR-SLMR and its robustness to complicated noise.

Xiaobo Wang, Shuo Wang, Shifeng Zhang et al.

Face recognition has witnessed significant progresses due to the advances of deep convolutional neural networks (CNNs), the central challenge of which, is feature discrimination. To address it, one group tries to exploit mining-based strategies (\textit{e.g.}, hard example mining and focal loss) to focus on the informative examples. The other group devotes to designing margin-based loss functions (\textit{e.g.}, angular, additive and additive angular margins) to increase the feature margin from the perspective of ground truth class. Both of them have been well-verified to learn discriminative features. However, they suffer from either the ambiguity of hard examples or the lack of discriminative power of other classes. In this paper, we design a novel loss function, namely support vector guided softmax loss (SV-Softmax), which adaptively emphasizes the mis-classified points (support vectors) to guide the discriminative features learning. So the developed SV-Softmax loss is able to eliminate the ambiguity of hard examples as well as absorb the discriminative power of other classes, and thus results in more discrimiantive features. To the best of our knowledge, this is the first attempt to inherit the advantages of mining-based and margin-based losses into one framework. Experimental results on several benchmarks have demonstrated the effectiveness of our approach over state-of-the-arts.