Jianxin Wu

AAAI Conference 2025 Conference Paper

All You Need in Knowledge Distillation Is a Tailored Coordinate System

• Junjie Zhou
• ke zhu
• Jianxin Wu

Knowledge Distillation (KD) is essential in transferring dark knowledge from a large teacher to a small student network, such that the student can be much more efficient than the teacher but with comparable accuracy. Existing KD methods, however, rely on a large teacher trained specifically for the target task, which is both very inflexible and inefficient. In this paper, we argue that a SSL-pretrained model can effectively act as the teacher and its dark knowledge can be captured by the coordinate system or linear subspace where the features lie in. We then need only one forward pass of the teacher, and then tailor the coordinate system (TCS) for the student network. Our TCS method is teacher-free and applies to diverse architectures, works well for KD and practical few-shot learning, allows cross-architecture distillation with large capacity gap. Experiments show that TCS achieves significantly higher accuracy than state-of-the-art KD methods, while only requiring roughly half of their training time and GPU memory costs.

NeurIPS Conference 2025 Conference Paper

GPLQ: A General, Practical, and Lightning QAT Method for Vision Transformers

• Guang Liang
• Xinyao Liu
• Jianxin Wu

Vision Transformers (ViTs) are essential in computer vision but are computationally intensive, too. Model quantization, particularly to low bit-widths like 4-bit, aims to alleviate this difficulty, yet existing Post-Training Quantization (PTQ) and Quantization-Aware Training (QAT) methods exhibit significant limitations. PTQ often incurs substantial accuracy drop, while QAT achieves high accuracy but suffers from prohibitive computational costs, limited generalization to downstream tasks, training instability, and lacking of open-source codebase. To address these challenges, this paper introduces General, Practical, and Lightning Quantization (GPLQ), a novel framework designed for efficient and effective ViT quantization. GPLQ is founded on two key empirical insights: the paramount importance of activation quantization and the necessity of preserving the model's original optimization basin to maintain generalization. Consequently, GPLQ employs a sequential activation-first, weights-later strategy. Stage 1 keeps weights in FP32 while quantizing activations with a feature mimicking loss in only 1 epoch to keep it stay in the same basin, thereby preserving generalization. Stage 2 quantizes weights using a PTQ method. As a result, GPLQ is 100x faster than existing QAT methods, lowers memory footprint to levels even below FP32 training, and achieves 4-bit model performance that is highly competitive with FP32 models in terms of both accuracy on ImageNet and generalization to diverse downstream tasks, including fine-grained visual classification and object detection. We release an easy-to-use open-source toolkit supporting multiple vision tasks at GPLQ code.

AAAI Conference 2025 Conference Paper

Treasures in Discarded Weights for LLM Quantization

• Hao Yu
• Yang Zhou
• Bohua Chen
• Zelan Yang
• Shen Li
• Yong Li
• Jianxin Wu

In recent years, large language models (LLMs) have developed rapidly and revolutionized natural language processing. However, high storage overhead and computing costs limit LLM deployment in resource-constrained environments. Quantization algorithms can effectively compress LLMs and accelerate inference, but they lead to loss in precision, especially in low-bit scenarios. In this paper, we find that the discarded weight values caused by quantization in fact contain treasures to improve LLMs' accuracy. To excavate those hidden treasures, we construct search spaces around these discarded weights and those weights within the search space can seamlessly be incorporated into the original quantization weights. To determine which weights should be merged, we design a plug-and-play weight compensation framework to capture global information and keep the weights with the highest potential benefits. Our framework can be combined with various LLM quantization algorithms to achieve higher precision without additional inference overhead. We validate the effectiveness of our approach on widely used benchmark datasets for LLMs.

NeurIPS Conference 2025 Conference Paper

Who Reasons in the Large Language Models?

• Jie Shao
• Jianxin Wu

Despite the impressive performance of large language models (LLMs), the process of endowing them with new capabilities---such as mathematical reasoning---remains largely empirical and opaque. A critical open question is whether reasoning abilities stem from the entire model, specific modules, or are merely artifacts of overfitting. In this work, we hypothesize that the reasoning capabilities in well-trained LLMs are primarily attributed to the output projection module (o proj) in the Transformer’s multi-head self-attention (MHSA) module. To support this hypothesis, we introduce Stethoscope for Networks (SfN), a suite of diagnostic tools designed to probe and analyze the internal behaviors of LLMs. Using SfN, we provide both circumstantial and empirical evidence suggesting that o proj plays a central role in enabling reasoning, whereas other modules contribute more to fluent dialogue. These findings offer a new perspective on LLM interpretability and open avenues for more targeted training strategies, potentially enabling more efficient and specialized LLMs.

NeurIPS Conference 2024 Conference Paper

DiffuLT: Diffusion for Long-tail Recognition Without External Knowledge

• Jie Shao
• ke zhu
• Hanxiao Zhang
• Jianxin Wu

This paper introduces a novel pipeline for long-tail (LT) recognition that diverges from conventional strategies. Instead, it leverages the long-tailed dataset itself to generate a balanced proxy dataset without utilizing external data or model. We deploy a diffusion model trained from scratch on only the long-tailed dataset to create this proxy and verify the effectiveness of the data produced. Our analysis identifies approximately-in-distribution (AID) samples, which slightly deviate from the real data distribution and incorporate a blend of class information, as the crucial samples for enhancing the generative model's performance in long-tail classification. We promote the generation of AID samples during the training of a generative model by utilizing a feature extractor to guide the process and filter out detrimental samples during generation. Our approach, termed Diffusion model for Long-Tail recognition (DiffuLT), represents a pioneer application of generative models in long-tail recognition. DiffuLT achieves state-of-the-art results on CIFAR10-LT, CIFAR100-LT, and ImageNet-LT, surpassing leading competitors by significant margins. Comprehensive ablations enhance the interpretability of our pipeline. Notably, the entire generative process is conducted without relying on external data or pre-trained model weights, which leads to its generalizability to real-world long-tailed scenarios.

AAAI Conference 2024 Conference Paper

DTL: Disentangled Transfer Learning for Visual Recognition

• Minghao Fu
• ke zhu
• Jianxin Wu

When pre-trained models become rapidly larger, the cost of fine-tuning on downstream tasks steadily increases, too. To economically fine-tune these models, parameter-efficient transfer learning (PETL) is proposed, which only tunes a tiny subset of trainable parameters to efficiently learn quality representations. However, current PETL methods are facing the dilemma that during training the GPU memory footprint is not effectively reduced as trainable parameters. PETL will likely fail, too, if the full fine-tuning encounters the out-of-GPU-memory issue. This phenomenon happens because trainable parameters from these methods are generally entangled with the backbone, such that a lot of intermediate states have to be stored in GPU memory for gradient propagation. To alleviate this problem, we introduce Disentangled Transfer Learning (DTL), which disentangles the trainable parameters from the backbone using a lightweight Compact Side Network (CSN). By progressively extracting task-specific information with a few low-rank linear mappings and appropriately adding the information back to the backbone, CSN effectively realizes knowledge transfer in various downstream tasks. We conducted extensive experiments to validate the effectiveness of our method. The proposed method not only reduces a large amount of GPU memory usage and trainable parameters, but also outperforms existing PETL methods by a significant margin in accuracy, achieving new state-of-the-art on several standard benchmarks.

AAAI Conference 2023 Conference Paper

Compressing Transformers: Features Are Low-Rank, but Weights Are Not!

• Hao Yu
• Jianxin Wu

Transformer and its variants achieve excellent results in various computer vision and natural language processing tasks, but high computational costs and reliance on large training datasets restrict their deployment in resource-constrained settings. Low-rank approximation of model weights has been effective in compressing CNN models, but its application to transformers has been less explored and is less effective. Existing methods require the complete dataset to fine-tune compressed models, which are both time-consuming and data-hungry. This paper reveals that the features (i.e., activations) are low-rank, but model weights are surprisingly not low-rank. Hence, AAFM is proposed, which adaptively determines the compressed model structure and locally compresses each linear layer's output features rather than the model weights. A second stage, GFM, optimizes the entire compressed network holistically. Both AAFM and GFM only use few training samples without labels, that is, they are few-shot, unsupervised, fast and effective. For example, with only 2K images without labels, 33% of the parameters are removed in DeiT-B with 18.8% relative throughput increase, but only a 0.23% accuracy loss for ImageNet recognition. The proposed methods are successfully applied to the language modeling task in NLP, too. Besides, the few-shot compressed models generalize well in downstream tasks.

AAAI Conference 2023 Conference Paper

Quantized Feature Distillation for Network Quantization

• ke zhu
• Yin-Yin He
• Jianxin Wu

Neural network quantization aims to accelerate and trim full-precision neural network models by using low bit approximations. Methods adopting the quantization aware training (QAT) paradigm have recently seen a rapid growth, but are often conceptually complicated. This paper proposes a novel and highly effective QAT method, quantized feature distillation (QFD). QFD first trains a quantized (or binarized) representation as the teacher, then quantize the network using knowledge distillation (KD). Quantitative results show that QFD is more flexible and effective (i.e., quantization friendly) than previous quantization methods. QFD surpasses existing methods by a noticeable margin on not only image classification but also object detection, albeit being much simpler. Furthermore, QFD quantizes ViT and Swin-Transformer on MS-COCO detection and segmentation, which verifies its potential in real world deployment. To the best of our knowledge, this is the first time that vision transformers have been quantized in object detection and image segmentation tasks.

AAAI Conference 2022 Conference Paper

A Random CNN Sees Objects: One Inductive Bias of CNN and Its Applications

• Yun-Hao Cao
• Jianxin Wu

This paper starts by revealing a surprising finding: without any learning, a randomly initialized CNN can localize objects surprisingly well. That is, a CNN has an inductive bias to naturally focus on objects, named as Tobias (“The object is at sight”) in this paper. This empirical inductive bias is further analyzed and successfully applied to self-supervised learning. A CNN is encouraged to learn representations that focus on the foreground object, by transforming every image into various versions with different backgrounds, where the foreground and background separation is guided by Tobias. Experimental results show that the proposed Tobias significantly improves downstream tasks, especially for object detection. This paper also shows that Tobias has consistent improvements on training sets of different sizes, and is more resilient to changes in image augmentations.

AAAI Conference 2021 Conference Paper

Bag of Tricks for Long-Tailed Visual Recognition with Deep Convolutional Neural Networks

• Yongshun Zhang
• Xiu-Shen Wei
• Boyan Zhou
• Jianxin Wu

In recent years, visual recognition on challenging long-tailed distributions, where classes often exhibit extremely imbalanced frequencies, has made great progress mostly based on various complex paradigms (e. g. , meta learning). Apart from these complex methods, simple refinements on training procedure also make contributions. These refinements, also called tricks, are minor but effective, such as adjustments in the data distribution or loss functions. However, different tricks might conflict with each other. If users apply these long-tail related tricks inappropriately, it could cause worse recognition accuracy than expected. Unfortunately, there has not been a scientific guideline of these tricks in the literature. In this paper, we first collect existing tricks in long-tailed visual recognition and then perform extensive and systematic experiments, in order to give a detailed experimental guideline and obtain an effective combination of these tricks. Furthermore, we also propose a novel data augmentation approach based on class activation maps for long-tailed recognition, which can be friendly combined with re-sampling methods and shows excellent results. By assembling these tricks scientifically, we can outperform state-of-the-art methods on four long-tailed benchmark datasets, including ImageNet-LT and iNaturalist 2018. Our code is open-source and available at https: //github. com/zhangyongshun/BagofTricks-LT.

AAAI Conference 2020 Conference Paper

Deep Discriminative CNN with Temporal Ensembling for Ambiguously-Labeled Image Classification

• Yao Yao
• Jiehui Deng
• Xiuhua Chen
• Chen Gong
• Jianxin Wu
• Jian Yang

In this paper, we study the problem of image classification where training images are ambiguously annotated with multiple candidate labels, among which only one is correct but is not accessible during the training phase. Due to the adopted non-deep framework and improper disambiguation strategies, traditional approaches are usually short of the representation ability and discrimination ability, so their performances are still to be improved. To remedy these two shortcomings, this paper proposes a novel approach termed “Deep Discriminative CNN” (D2 CNN) with temporal ensembling. Specifically, to improve the representation ability, we innovatively employ the deep convolutional neural networks for ambiguously-labeled image classification, in which the wellknown ResNet is adopted as our backbone. To enhance the discrimination ability, we design an entropy-based regularizer to maximize the margin between the potentially correct label and the unlikely ones of each image. In addition, we utilize the temporally assembled predictions of different epochs to guide the training process so that the latent groundtruth label can be confidently highlighted. This is much superior to the traditional disambiguation operations which treat all candidate labels equally and identify the hidden groundtruth label via some heuristic ways. Thorough experimental results on multiple datasets firmly demonstrate the effectiveness of our proposed D2 CNN when compared with other existing stateof-the-art approaches.

AAAI Conference 2020 Conference Paper

Repetitive Reprediction Deep Decipher for Semi-Supervised Learning

• Guo-Hua Wang
• Jianxin Wu

Most recent semi-supervised deep learning (deep SSL) methods used a similar paradigm: use network predictions to update pseudo-labels and use pseudo-labels to update network parameters iteratively. However, they lack theoretical support and cannot explain why predictions are good candidates for pseudo-labels. In this paper, we propose a principled end-toend framework named deep decipher (D2) for SSL. Within the D2 framework, we prove that pseudo-labels are related to network predictions by an exponential link function, which gives a theoretical support for using predictions as pseudo-labels. Furthermore, we demonstrate that updating pseudo-labels by network predictions will make them uncertain. To mitigate this problem, we propose a training strategy called repetitive reprediction (R2). Finally, the proposed R2-D2 method is tested on the large-scale ImageNet dataset and outperforms state-of-the-art methods by 5 percentage points.

AAAI Conference 2018 Conference Paper

Action Recognition With Coarse-to-Fine Deep Feature Integration and Asynchronous Fusion

• Weiyao Lin
• Chongyang Zhang
• Ke Lu
• Bin Sheng
• Jianxin Wu
• Bingbing Ni
• Xin Liu
• Hongkai Xiong

Action recognition is an important yet challenging task in computer vision. In this paper, we propose a novel deepbased framework for action recognition, which improves the recognition accuracy by: 1) deriving more precise features for representing actions, and 2) reducing the asynchrony between different information streams. We first introduce a coarse-to-fine network which extracts shared deep features at different action class granularities and progressively integrates them to obtain a more accurate feature representation for input actions. We further introduce an asynchronous fusion network. It fuses information from different streams by asynchronously integrating stream-wise features at different time points, hence better leveraging the complementary information in different streams. Experimental results on action recognition benchmarks demonstrate that our approach achieves the state-of-the-art performance.

IJCAI Conference 2018 Conference Paper

Age Estimation Using Expectation of Label Distribution Learning

• Bin-Bin Gao
• Hong-Yu Zhou
• Jianxin Wu
• Xin Geng

Age estimation performance has been greatly improved by using convolutional neural network. However, existing methods have an inconsistency between the training objectives and evaluation metric, so they may be suboptimal. In addition, these methods always adopt image classification or face recognition models with a large amount of parameters, which bring expensive computation cost and storage overhead. To alleviate these issues, we design a lightweight network architecture and propose a unified framework which can jointly learn age distribution and regress age. The effectiveness of our approach has been demonstrated on apparent and real age estimation tasks. Our method achieves new state-of-the-art results using the single model with 36$\times$ fewer parameters and 2. 6$\times$ reduction in inference time. Moreover, our method can achieve comparable results as the state-of-the-art even though model parameters are further reduced to 0. 9M~(3. 8MB disk storage). We also analyze that Ranking methods are implicitly learning label distributions.

IJCAI Conference 2017 Conference Paper

Deep Descriptor Transforming for Image Co-Localization

• Xiu-Shen Wei
• Chen-Lin Zhang
• Yao Li
• Chen-Wei Xie
• Jianxin Wu
• Chunhua Shen
• Zhi-Hua Zhou

Reusable model design becomes desirable with the rapid expansion of machine learning applications. In this paper, we focus on the reusability of pre-trained deep convolutional models. Specifically, different from treating pre-trained models as feature extractors, we reveal more treasures beneath convolutional layers, i. e. , the convolutional activations could act as a detector for the common object in the image co-localization problem. We propose a simple but effective method, named Deep Descriptor Transforming (DDT), for evaluating the correlations of descriptors and then obtaining the category-consistent regions, which can accurately locate the common object in a set of images. Empirical studies validate the effectiveness of the proposed DDT method. On benchmark image co-localization datasets, DDT consistently outperforms existing state-of-the-art methods by a large margin. Moreover, DDT also demonstrates good generalization ability for unseen categories and robustness for dealing with noisy data.

AAAI Conference 2016 Conference Paper

Representing Sets of Instances for Visual Recognition

• Jianxin Wu
• Bin-Bin Gao
• Guoqing Liu

In computer vision, a complex entity such as an image or video is often represented as a set of instance vectors, which are extracted from different parts of that entity. Thus, it is essential to design a representation to encode information in a set of instances robustly. Existing methods such as FV and VLAD are designed based on a generative perspective, and their performances fluctuate when difference types of instance vectors are used (i. e. , they are not robust). The proposed D3 method effectively compares two sets as two distributions, and proposes a directional total variation distance (DTVD) to measure their dissimilarity. Furthermore, a robust classifier-based method is proposed to estimate DTVD robustly, and to efficiently represent these sets. D3 is evaluated in action and image recognition tasks. It achieves excellent robustness, accuracy and speed.

IJCAI Conference 2013 Conference Paper

Reduced Heteroscedasticity Linear Regression for Nyström Approximation

• Hao Yang
• Jianxin Wu

The Nyström method is a well known sampling based low-rank matrix approximation approach. It is usually considered to be originated from the numerical treatment of integral equations and eigendecomposition of matrices. In this paper, we present a novel point of view for the Nyström approximation. We show that theoretically the Nyström method can be regraded as a set of pointwise ordinary least square linear regressions of the kernel matrix, sharing the same design matrix. With the new interpretation, we are able to analyze the approximation quality based on the fulfillment of the homoscedasticity assumption and explain the success and deficiency of various sampling methods. We also empirically show that positively skewed explanatory variable distributions can lead to heteroscedasticity. Based on this discovery, we propose to use non-symmetric explanatory functions to improve the quality of the Nyström approximation with almost no extra computational cost. Experiments show that positively skewed datasets widely exist, and our method exhibits good improvements on these datasets.

JMLR Journal 2011 Journal Article

Efficient and Effective Visual Codebook Generation Using Additive Kernels

• Jianxin Wu
• Wei-Chian Tan
• James M. Rehg

Common visual codebook generation methods used in a bag of visual words model, for example, k-means or Gaussian Mixture Model, use the Euclidean distance to cluster features into visual code words. However, most popular visual descriptors are histograms of image measurements. It has been shown that with histogram features, the Histogram Intersection Kernel (HIK) is more effective than the Euclidean distance in supervised learning tasks. In this paper, we demonstrate that HIK can be used in an unsupervised manner to significantly improve the generation of visual codebooks. We propose a histogram kernel k-means algorithm which is easy to implement and runs almost as fast as the standard k-means. The HIK codebooks have consistently higher recognition accuracy over k-means codebooks by 2-4% in several benchmark object and scene recognition data sets. The algorithm is also generalized to arbitrary additive kernels. Its speed is thousands of times faster than a naive implementation of the kernel k-means algorithm. In addition, we propose a one-class SVM formulation to create more effective visual code words. Finally, we show that the standard k-median clustering method can be used for visual codebook generation and can act as a compromise between the HIK / additive kernel and the k-means approaches. [abs] [ pdf ][ bib ] &copy JMLR 2011. ( edit, beta )

IJCAI Conference 2011 Conference Paper

Probit Classifiers with a Generalized Gaussian Scale Mixture Prior

• Guoqing Liu
• Jianxin Wu
• Suiping Zhou

Most of the existing probit classifiers are based on sparsity-oriented modeling. However, we show that sparsity is not always desirable in practice, and only an appropriate degree of sparsity is profitable. In this work, we propose a flexible probabilistic model using a generalized Gaussian scale mixture prior that can promote an appropriate degree of sparsity for its model parameters, and yield either sparse or non-sparse estimates according to the intrinsic sparsity of features in a dataset. Model learning is carried out by an efficient modified maximum a posteriori (MAP) estimate. We also show relationships of the proposed model to existing probit classifiers as well as iteratively re-weighted l1 and l2 minimizations. Experiments demonstrate that the proposed method has better or comparable performances in feature selection for linear classifiers as well as in kernel-based classification.

AIJ Journal 2010 Journal Article

Corrigendum to “Ensembling neural networks: Many could be better than all” [Artificial Intelligence 137 (1–2) (2002) 239–263]

• Zhi-Hua Zhou
• Jianxin Wu
• Wei Tang

ICRA Conference 2009 Conference Paper

On the switching control of multiple mobile robots formation

• Jianxin Wu
• Zhe Jiang

This paper considers the formation control of multiple skid-steered mobile robots with nonholonomic constraint in each robot dynamics. We propose a switching control method based on local sensor-based information for the group of robots. If the distance between leader and follower robots is larger than the switching threshold, a nonlinear control law is employed to make the followers move effectively toward the leader robot; otherwise, finite-time control based on feedback linearization is used to quickly stabilize the relative distance and orientation of the follower robots. Both stability analysis and simulations are provided and moreover, the formation control of Pioneer 3-AT mobile robots by the proposed method is shown.

NeurIPS Conference 2003 Conference Paper

Learning a Rare Event Detection Cascade by Direct Feature Selection

• Jianxin Wu
• James Rehg
• Matthew Mullin

Face detection is a canonical example of a rare event detection prob- lem, in which target patterns occur with much lower frequency than non- targets. Out of millions of face-sized windows in an input image, for ex- ample, only a few will typically contain a face. Viola and Jones recently proposed a cascade architecture for face detection which successfully ad- dresses the rare event nature of the task. A central part of their method is a feature selection algorithm based on AdaBoost. We present a novel cascade learning algorithm based on forward feature selection which is two orders of magnitude faster than the Viola-Jones approach and yields classifiers of equivalent quality. This faster method could be used for more demanding classification tasks, such as on-line learning.

AIJ Journal 2002 Journal Article

Ensembling neural networks: Many could be better than all

• Zhi-Hua Zhou
• Jianxin Wu
• Wei Tang

Neural network ensemble is a learning paradigm where many neural networks are jointly used to solve a problem. In this paper, the relationship between the ensemble and its component neural networks is analyzed from the context of both regression and classification, which reveals that it may be better to ensemble many instead of all of the neural networks at hand. This result is interesting because at present, most approaches ensemble all the available neural networks for prediction. Then, in order to show that the appropriate neural networks for composing an ensemble can be effectively selected from a set of available neural networks, an approach named GASEN is presented. GASEN trains a number of neural networks at first. Then it assigns random weights to those networks and employs genetic algorithm to evolve the weights so that they can characterize to some extent the fitness of the neural networks in constituting an ensemble. Finally it selects some neural networks based on the evolved weights to make up the ensemble. A large empirical study shows that, compared with some popular ensemble approaches such as Bagging and Boosting, GASEN can generate neural network ensembles with far smaller sizes but stronger generalization ability. Furthermore, in order to understand the working mechanism of GASEN, the bias-variance decomposition of the error is provided in this paper, which shows that the success of GASEN may lie in that it can significantly reduce the bias as well as the variance.