The 2nd Workshop on
Artificial Intelligence for Anomalies and Novelties
in Conjunction with IJCAI 2021
The 2nd Workshop on Artificial Intelligence for Anomalies and Novelties (AI4AN 2021) will be co-located with IJCAI 2021 and held online on 20-21 August 2021, including the following two sessions:
1st Session, Aug 20 20:00 – 2:00 (Next day) Montreal Time (UTC-4)
2nd Session, Aug 21 10:00 – 16:00 Montreal Time (UTC-4)
November 03, 2021: The workshop videos are now released and available at https://ijcai-21.org/video-page/?video=W20
August 22, 2021: The recording video of the workshop will be released soon. Stay tuned!
August 14, 2021: The program will feature an exciting line-up of 6 keynote talks, 3 invited talks, and 3 interesting accepted papers. Stay tuned!
March 25, 2021: AI4AN 2021 workshop website is up!
Submission Due: June 1, 2021 (23:59 UTC-12)
Notification Due: July 1, 2021 (23:59 UTC-12)
Camera-ready Due: July 14, 2021 (23:59 UTC-12)
Anomalies are referred to as observations or events that are rare or significantly different from the majority of observations we have in hand, while novelties are observations from novel classes that were unseen during learning. Recognition, detection and accommodation to anomalies and novelties are active research areas in multiple communities, including data mining, machine learning, and computer vision. Some of the most relevant well-established research areas include anomaly detection, out-of-distribution example detection, adversarial example recognition and detection, curiosity-driven reinforcement learning, open-set recognition and adaptation. The successful early detection of anomalies and novelties is of great significance across many domains. For example, it may prevent the loss of billions of dollars by its application to fraud detection and anti-money laundering in fintech, save millions of lives through early disease detection, safeguard large-scale computer networks and data centers from malicious attacks by its use in intrusion detection, defend AI systems from adversarial attacks, and equip AI systems with the ability to work safely in open worlds. Specialized techniques have been studied in some of these areas for decades, but recent developments are raising a wide variety of new research questions. First, anomaly detection in deep learning is challenging, because the learned latent representations—while they are sufficient for accurate performance on nominal inputs—often fail to represent anomalies and novelties in a way that allows them to be detected. Second, there is a need to create fundamental theories of novelty that can articulate what anomalies can be detected and how much data and computation is required. Third, detection is only the first step in enabling an AI system to adapt successfully to novelties. AI systems need to be able to characterize the nature of the novelties and then develop both short-term and long-term responses to accommodating them.
This workshop will gather researchers and practitioners from diverse communities and knowledge background to promote the development of fundamental theories, effective algorithms, and novel applications of anomaly and novelty detection, characterization, and adaptation.