Author name cluster

Junwei Chen

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

4 papers

2 author rows

JBHI Journal 2025 Journal Article

GRATCR: Epitope-Specific T Cell Receptor Sequence Generation With Data-Efficient Pre-Trained Models

Zhenghong Zhou
Junwei Chen
Shenggeng Lin
Liang Hong
Dong-Qing Wei
Yi Xiong

T cell receptors (TCRs) play a crucial role in numerous immunotherapies targeting tumor cells. However, their acquisition and optimization present significant challenges, involving laborious and time-consuming wet lab experimental resource. Deep generative models have demonstrated remarkable capabilities in functional protein sequence generation, offering a promising solution for enhancing the acquisition of specific TCR sequences. Here, we propose GRATCR, a framework incorporates two pre-trained modules through a novel “grafting” strategy, to de-novo generate TCR sequences targeting specific epitopes. Experimental results demonstrate that TCRs generated by GRATCR exhibit higher specificity toward desired epitopes and are more biologically functional compared with the state-of-the-art model, by using significantly fewer training data. Additionally, the generated sequences display novelty compared to natural sequences, and the interpretability evaluation further confirmed that the model is capable of capturing important binding patterns.

Details DOI

ECAI Conference 2025 Conference Paper

Training Robotic Self-Evolving with GRPO

Qinpeng Yi
Ping Zhang
Junwei Chen

Current embodied robots heavily depend on pre-trained models, whose capabilities are inherently constrained by the data they were originally trained on. However, truly intelligent robots are expected to improve themselves autonomously when encountering novel environments where these pre-trained models fall short. This is the capability we define as self-evolving ability. In this paper, we investigate the self-evolving capacity of robotic vision models. Specifically, we simulate this process using the R3ED dataset and propose a training framework in which a policy learns to navigate through unfamiliar environments to collect informative data that can be used to refine the vision model. Our training pipeline is built upon the GRPO algorithm and incorporates historical states into the policy design to enhance contextual awareness. Furthermore, we introduce a novel reward mechanism based on supervision discrepancy to guide effective data collection. Experimental results validate the effectiveness of our proposed reinforcement training strategy. Our work highlights the potential of designing intelligent robots that can improve themselves without the intervene of human beings. Nevertheless, we acknowledge that robotic self-evolving remains a nascent and underexplored area, with significant room for further future research and the discovery of more optimal approaches.

Details

ICML Conference 2024 Conference Paper

Harnessing Neural Unit Dynamics for Effective and Scalable Class-Incremental Learning

Depeng Li 0001
Tianqi Wang
Junwei Chen
Wei Dai 0004
Zhigang Zeng

Class-incremental learning (CIL) aims to train a model to learn new classes from non-stationary data streams without forgetting old ones. In this paper, we propose a new kind of connectionist model by tailoring neural unit dynamics that adapt the behavior of neural networks for CIL. In each training session, it introduces a supervisory mechanism to guide network expansion whose growth size is compactly commensurate with the intrinsic complexity of a newly arriving task. This constructs a near-minimal network while allowing the model to expand its capacity when cannot sufficiently hold new classes. At inference time, it automatically reactivates the required neural units to retrieve knowledge and leaves the remaining inactivated to prevent interference. We name our model AutoActivator, which is effective and scalable. To gain insights into the neural unit dynamics, we theoretically analyze the model’s convergence property via a universal approximation theorem on learning sequential mappings, which is under-explored in the CIL community. Experiments show that our method achieves strong CIL performance in rehearsal-free and minimal-expansion settings with different backbones.

Details

AAAI Conference 2024 Conference Paper

Towards Continual Learning Desiderata via HSIC-Bottleneck Orthogonalization and Equiangular Embedding

Depeng Li
Tianqi Wang
Junwei Chen
Qining Ren
Kenji Kawaguchi
Zhigang Zeng

Deep neural networks are susceptible to catastrophic forgetting when trained on sequential tasks. Various continual learning (CL) methods often rely on exemplar buffers or/and network expansion for balancing model stability and plasticity, which, however, compromises their practical value due to privacy and memory concerns. Instead, this paper considers a strict yet realistic setting, where the training data from previous tasks is unavailable and the model size remains relatively constant during sequential training. To achieve such desiderata, we propose a conceptually simple yet effective method that attributes forgetting to layer-wise parameter overwriting and the resulting decision boundary distortion. This is achieved by the synergy between two key components: HSIC-Bottleneck Orthogonalization (HBO) implements non-overwritten parameter updates mediated by Hilbert-Schmidt independence criterion in an orthogonal space and EquiAngular Embedding (EAE) enhances decision boundary adaptation between old and new tasks with predefined basis vectors. Extensive experiments demonstrate that our method achieves competitive accuracy performance, even with absolute superiority of zero exemplar buffer and 1.02x the base model.

PDF Details DOI