3rd International Workshop on Generalizing from
Limited Resources in the Open World

Workshop at International Joint Conference on Artificial Intelligence (IJCAI) 2025

Overview

The capacity for flexible and efficient generalization from limited resources is a hallmark of human intelligence. However, artificial intelligence (AI) models often encounter challenges in extending their capabilities from limited resources like constrained training data and computation, particularly in fields such as computer vision and natural language processing. The struggle to generalize effectively from limited resources hinders the practical deployment of AI models in the open world, where efficient learning from sparse samples in unpredictable scenarios is essential. Recognizing this imperative, our workshop aims to delve into the intricate landscape of AI model generalization in open-world settings to foster advancements in real-world AI applications. Numerous endeavors have been undertaken to surmount these challenges, addressing the inefficiencies in learning models and bridging the divide between human and machine intelligence.

This workshop centers on the academic exploration of efficient methodologies within the realm of artificial intelligence (AI) models. Our focus encompasses both data-efficient strategies, such as zero/few-shot learning and domain adaptation, as well as model-efficient approaches like model sparsification and compact model design. By convening researchers specializing in these areas, our objective is to facilitate the sharing of recent research findings and engage in discussions about the future trajectories of AI model generalization. The significance of this workshop is underscored by the timely relevance of its thematic focus, which has garnered substantial attention from the research community due to its direct implications for practical applications. Through this forum, we aspire to foster the exchange of ideas, providing an avenue for the articulation of novel insights aimed at addressing the generalization challenges confronting AI models, thereby contributing to the advancement of real-world AI applications.

Call for Papers

We invite submissions on any aspect of generalizing from limited resources in the open world. Considering the recent significant success of large model, in this year, we will include more generalization approaches in openworld for generative AI models. We welcome research contributions related to the following (but not limited to) topics:

New methods for in-context learning
Applications of large AI models in vertical domains
New methods for AI alignment
New methods and benchmarks of open set/world learning problem
New methods for few-/zero-shot learning
New methods for domain-adaptation methods
New methods for training generative models under limited data
Benchmark for evaluating model generalization
Understanding the generalization vulnerabilities of deep learning systems
Network sparsity, quantization, distillation, etc.
Neural architecture search (NAS)
Efficient network architecture design
Efficient methods for generative models like diffusion, large language models
Hardware implementation and on-device deployment
On-device learning
Brain-inspired artificial intelligence like spiking neural networks (SNN)
Optimization on parallel and distributed training
New methods for robotics
New method for embodied intelligence

Submission Format: Submissions need to be anonymized and follow Springer's guidelines and author instructions. The workshop considers two types of submissions: (1) Full Paper: Papers are limited to 12-15 pages; (2) Short Paper: Papers are limited to 4-11 pages.
Peer review: Paper submissions must conform with the “double-blind” review policy. All papers will be peer-reviewed by experts in the field, they will receive at least two reviews. Based on the area chair recommendations, the accepted papers will be allocated either a contributed talk or a poster presentation.
Important: The accepted papers will be published on the proceeding of Communications in Computer and Information Science, indexed in EI-Compendex, etc. Additionally, Best Paper and other recognitions will be awarded.
Submission Site: https://openreview.net/group?id=ijcai.org/IJCAI/2025/Workshop/GLOW
Submission Deadline: May 30, 2025 (11: 59PM UTC-0)

Speakers

Berkin Akin

DeepMind

Frank Hutter

University of Freiburg

Tim Dettmers

Carnegie Mellon University

Liang Zheng

Australian National University

Huanrui Yang

The University of Arizona

Yue He

Renmin University

Schedule

Room TBD

Challenge

Based on our published X-ray open-world prohibited item detection datasets, we also bring a challenge to encourage participants to use sparse training data to design models capable of effectively identifying known and unknown prohibited objects, promoting the protection of public transport safety in society. This challenge devises a track for participants: rotated object detection for X-ray prohibited items according to practical demand. Welcome to join the challenge at our platform!

Rotated X-ray Prohibited Items Detection

Horizontal bounding boxes are incapable of representing slender prohibited items in various orientations, which involve in massive background information. Therefore exploring rotated object detection task on X-ray security inspection scenario is significant.

To accelerate the research on enhancing rotated object detection performance in the X-ray scenario, we organize this challenge track, where image data in real X-ray security inspection scene with rotated bounding boxes are provided.