Author name cluster

Xinjun Mao

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

AAAI Conference 2025 Conference Paper

Maintaining Fairness in Logit-based Knowledge Distillation for Class-Incremental Learning

Zijian Gao
Shanhao Han
Xingxing Zhang
Kele Xu
Dulan Zhou
Xinjun Mao
Yong Dou
Huaimin Wang

Logit-based knowledge distillation (KD) is commonly used to mitigate catastrophic forgetting in class-incremental learning (CIL) caused by data distribution shifts. However, the strict match of logit values between student and teacher models conflicts with the cross-entropy (CE) loss objective of learning new classes, leading to significant recency bias (i.e. unfairness). To address this issue, we rethink the overlooked limitations of KD-based methods through empirical analysis. Inspired by our findings, we introduce a plug-and-play pre-process method that normalizes the logits of both the student and teacher across all classes, rather than just the old classes, before distillation. This approach allows the student to focus on both old and new classes, capturing intrinsic inter-class relations from the teacher. By doing so, our method avoids the inherent conflict between KD and CE, maintaining fairness between old and new classes. Additionally, recognizing that overconfident teacher predictions can hinder the transfer of inter-class relations (i.e., dark knowledge), we extend our method to capture intra-class relations among different instances, ensuring fairness within old classes. Our method integrates seamlessly with existing logit-based KD approaches, consistently enhancing their performance across multiple CIL benchmarks without incurring additional training costs.

PDF Details DOI

NeurIPS Conference 2024 Conference Paper

Stabilizing Zero-Shot Prediction: A Novel Antidote to Forgetting in Continual Vision-Language Tasks

Zijian Gao
Xingxing Zhang
Kele Xu
Xinjun Mao
Huaimin Wang

Continual learning (CL) empowers pre-trained vision-language (VL) models to efficiently adapt to a sequence of downstream tasks. However, these models often encounter challenges in retaining previously acquired skills due to parameter shifts and limited access to historical data. In response, recent efforts focus on devising specific frameworks and various replay strategies, striving for a typical learning-forgetting trade-off. Surprisingly, both our empirical research and theoretical analysis demonstrate that the stability of the model in consecutive zero-shot predictions serves as a reliable indicator of its anti-forgetting capabilities for previously learned tasks. Motivated by these insights, we develop a novel replay-free CL method named ZAF (Zero-shot Antidote to Forgetting), which preserves acquired knowledge through a zero-shot stability regularization applied to wild data in a plug-and-play manner. To enhance efficiency in adapting to new tasks and seamlessly access historical models, we introduce a parameter-efficient EMA-LoRA neural architecture based on the Exponential Moving Average (EMA). ZAF utilizes new data for low-rank adaptation (LoRA), complemented by a zero-shot antidote on wild data, effectively decoupling learning from forgetting. Our extensive experiments demonstrate ZAF's superior performance and robustness in pre-trained models across various continual VL concept learning tasks, achieving leads of up to 3. 70\%, 4. 82\%, and 4. 38\%, along with at least a 10x acceleration in training speed on three benchmarks, respectively. Additionally, our zero-shot antidote significantly reduces forgetting in existing models by at least 6. 37\%. Our code is available at https: //github. com/Zi-Jian-Gao/Stabilizing-Zero-Shot-Prediction-ZAF.

PDF Details DOI

ICRA Conference 2021 Conference Paper

Towards Adjoint Sensing and Acting Schemes and Interleaving Task Planning for Robust Robot Plan

Shuo Yang 0005
Xinjun Mao
Shuo Wang
Huaiyu Xiao
Yuanzhou Xue

Robots operating in open environments expect to have robust plans to achieve tasks successfully under environment uncertainties. However, both partial observability and dynamics of environment states have significantly decreased the robustness of task achievement, making robot task planning much more challenging. The partially observable states require the robot to obtain observations for optimally acting of the task goal. Also, state dynamics expects the robot to continuously observe surroundings for acting safely. Both challenges practically demand the purposeful and tight interactions between robot state-changing actuating actions and sensor-based observation actions. This paper proposes a novel model of Adjoint Sensing and Acting (ASA) that explicitly defines two parallel and sequential interaction schemes between actuating and observation actions, as well as an extended Behavior Tree for a concrete implementation of above schemes. We further propose an interleaving task planning approach for planning ASA-style plans, which integrates a deliberative POMDP planner for pursuing task goals, and a reactive Behavior Tree executive for fast responding to unexpected events. We experimentally demonstrate that ASA interaction schemes are practical and applicable to model and plan the open environment robot tasks. The plans from the interleaving task planning approach are both reactive in run-time response and efficient in task achievement.

Details

AAMAS Conference 2012 Conference Paper

OrgMAP: An Organization-based Approach for Multi-Agent Programming

Cuiyun Hu
Xinjun Mao
Yin Chen
Huiping Zhou

This paper proposes a new organization-based multi-agent programming (OrgMAP) approach to constructing dynamic and flexible software systems. A computational and programming model named Oragent is defined following software engineering principles such as modularity, reusability and etc. Oragent model not only allows programmers to represent the systems with highlevel abstractions in terms of organizations, rules, protocols and roles, but also provides a number of mechanisms, such as encapsulation, inheritance, enactment and event, to improve the dynamics and flexibility of MAS.

PDF