Continual Learning of a Mixed Sequence of Similar and Dissimilar Tasks

By Zixuan Ke, Bing Liu, Xingchang Huang. 2021

165 citations recorded by the source corpus.

Existing research on continual learning of a sequence of tasks focused on dealing with catastrophic forgetting, where the tasks are assumed to be dissimilar and have little shared knowledge. Some work has also been done to transfer previously learned knowledge to the new task when the tasks are similar and have shared knowledge. To the best of our knowledge, no technique has been proposed to learn a sequence of mixed similar and dissimilar tasks that can deal with forgetting and also transfer knowledge forward and backward. This paper proposes such a technique to learn both types of tasks in the same network. For dissimilar tasks, the algorithm focuses on dealing with forgetting, and for similar tasks, the algorithm focuses on selectively transferring the knowledge learned from some similar previous tasks to improve the new task learning. Additionally, the algorithm automatically detects whether a new task is similar to any previous tasks. Empirical evaluation using sequences of mixed tasks demonstrates the effectiveness of the proposed model.

Subfields: Artificial Intelligence, Computational Mechanics, Computer Vision and Pattern Recognition, Meta & Continual Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computational Mechanics

In Computer Vision and Pattern Recognition

In Meta & Continual Learning

Citations in the corpus

A Comprehensive Survey of Continual Learning: Theory, Method and Application (2023, background): “…realistic consider-ationshavebeenincorporated,suchasmultiplelabels [29],noisy labels [30], blurred boundary [31], [32], hierarchical granularity [33] and sub-populations [34], mixture of task similarity [35], long-tailed distribution [36], novel class discovery [37], [38], multi-modality [39], etc.”
Learnability and Algorithm for Continual Learning (2023, background): “Several techniques have also been proposed to carry out knowledge transfer (Ke et al., 2020; 2021; Lin et al., 2022).”
A Theoretical Study on Solving Continual Learning (2022, background): “HAT [3] and CAT [52] protect previous tasks by masking the important parameters to those tasks.”
Beyond Not-Forgetting: Continual Learning with Backward Knowledge Transfer (2022, background): “[11] studies the knowledge transfer in a mixed sequence of similar and dissimilar tasks, but employs a complicated task similarity detection mechanism where two additional networks need to be trained first for each task before learning the new task.”
On the Effectiveness of Lipschitz-Driven Rehearsal in Continual Learning (2022, neutral): “Differently, structural methods purposefully organize the allocation of model capacity to prevent interference and facilitate parameter sharing [1, 51, 66, 37].”
Adaptive Orthogonal Projection for Batch and Online Continual Learning (2022, background): “Several works have performed knowledge transfer across tasks (Ke, Liu, and Huang 2020; Ke et al. 2021; Schwarz et al. 2018; Fernando et al. 2017; Rusu et al. 2016).”
DualPrompt: Complementary Prompting for Rehearsal-free Continual Learning (2022, methodology): “These methods can be further categorized as expanding the model [51, 65, 26, 31, 59], or dividing the model [35, 52, 58, 17, 11].”
Achieving Forgetting Prevention and Knowledge Transfer in Continual Learning (2021, methodology): “From the 10 systems, we created 10 baselines using W2V embeddings with the aspect term added before the sentence so that the Task-CL methods can take both aspect and the review sentence for ASC; 7 baselines using Adapter-BERT (SRK, KAN and CAT’s architectures cannot work with adapters); and 10 basel”
Learngene: From Open-World to Your Learning Task (2021, methodology): “There are three mainstream methods in deep learning including regularization approaches [38, 29, 32, 72, 31], dynamic-architectures approaches [50, 69, 14, 49], and memory-replay approaches [57, 54, 40, 31, 30].”
Adapting BERT for Continual Learning of a Sequence of Aspect Sentiment Classification Tasks (2021, background): “Ke et al. (2020a) also performed transfer in the image domain.”
Continual Learning by Using Information of Each Class Holistically (2021, methodology): “CAT is able to perform knowledge transfer.”
Online Continual Learning through Mutual Information Maximization (2022, methodology): “As an extension, CAT (Ke et al., 2020) uses a parameter isolation approach to deal with both CF and knowledge transfer across tasks.”
How to Reuse and Compose Knowledge for a Lifetime of Tasks: A Survey on Continual Learning and Functional Composition (2022, methodology): “Another technique is to identify the most similar tasks by training one model via transfer and another via STL and comparing their validation performances (Ke et al., 2020).”
Consistent Prompting for Rehearsal-Free Continual Learning (2024, unknown): “Parameter isolation methods [13, 20, 48] freeze specific parameters while allocating the rest for subsequent tasks or expanding the network for new knowledge learning.”