Overcoming Catastrophic Forgetting in Continual Learning by Exploring Eigenvalues of Hessian Matrix

Kong, Yajing; Liu, Liu; Chen, Huanhuan; Kacprzyk, Janusz; Tao, Dacheng

doi:10.1109/tnnls.2023.3292359

Cited by 6 publications

(2 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, the regularization-based CL method constructs a regularization-based term as an approximate loss function of previous tasks to limit the updating of parameters, which may cause the network parameters to not reflect the current new tasks in a timely manner. Most of the methods are based on heuristics, with no good theoretical understanding of the factors associated with catastrophic forgetting [47].…”

Section: Continuous Learning Methods Based On Regularizationmentioning

confidence: 99%

CL-BPUWM: continuous learning with Bayesian parameter updating and weight memory

He,

Yang,

et al. 2024

Complex Intell. Syst.

View full text Add to dashboard Cite

Catastrophic forgetting in neural networks is a common problem, in which neural networks lose information from previous tasks after training on new tasks. Although adopting a regularization method that preferentially retains the parameters important to the previous task to avoid catastrophic forgetting has a positive effect; existing regularization methods cause the gradient to be near zero because the loss is at the local minimum. To solve this problem, we propose a new continuous learning method with Bayesian parameter updating and weight memory (CL-BPUWM). First, a parameter updating method based on the Bayes criterion is proposed to allow the neural network to gradually obtain new knowledge. The diagonal of the Fisher information matrix is then introduced to significantly minimize computation and increase parameter updating efficiency. Second, we suggest calculating the importance weight by observing how changes in each network parameter affect the model prediction output. In the process of model parameter updating, the Fisher information matrix and the sensitivity of the network are used as the quadratic penalty terms of the loss function. Finally, we apply dropout regularization to reduce model overfitting during training and to improve model generalizability. CL-BPUWM performs very well in continuous learning for classification tasks on CIFAR-100 dataset, CIFAR-10 dataset, and MNIST dataset. On CIFAR-100 dataset, it is 0.8%, 1.03% and 0.75% higher than the best performing regularization method (EWC) in three task partitions. On CIFAR-10 dataset, it is 2.25% higher than the regularization method (EWC) and 0.7% higher than the scaled method (GR). It is 0.66% higher than the regularization method (EWC) on the MNIST dataset. When the CL-BPUWM method was combined with the brain-inspired replay model under the CIFAR-100 and CIFAR-10 datasets, the classification accuracy was 2.35% and 5.38% higher than that of the baseline method, BI-R + SI.

show abstract

Section: Continuous Learning Methods Based On Regularizationmentioning

confidence: 99%

CL-BPUWM: continuous learning with Bayesian parameter updating and weight memory

He,

Yang,

et al. 2024

Complex Intell. Syst.

View full text Add to dashboard Cite

show abstract

“…This means they may face two potential problems: catastrophic forgetting (losing previous knowledge and skills) or interference (degrading new knowledge and skills) [162]. The brain can prevent forgetting and interference through sleep, rehearsal, and consolidation [159], while continuous learning techniques use methods such as regularization, replay, and distillation to mitigate them [171].…”

mentioning

confidence: 99%

A Perspective on Prosthetic Hands Control: From the Brain to the Hand

Gentile,

Gruppioni

2023

Prosthesis

View full text Add to dashboard Cite

The human hand is a complex and versatile organ that enables humans to interact with the environment, communicate, create, and use tools. The control of the hand by the brain is a crucial aspect of human cognition and behaviour, but also a challenging problem for both neuroscience and engineering. The aim of this study is to review the current state of the art in hand and grasp control from a neuroscientific perspective, focusing on the brain mechanisms that underlie sensory integration for hand control and the engineering implications for developing artificial hands that can mimic and interface with the human brain. The brain controls the hand by processing and integrating sensory information from vision, proprioception, and touch, using different neural pathways. The user’s intention can be obtained to control the artificial hand by using different interfaces, such as electromyography, electroneurography, and electroencephalography. This and other sensory information can be exploited by different learning mechanisms that can help the user adapt to changes in sensory inputs or outputs, such as reinforcement learning, motor adaptation, and internal models. This work summarizes the main findings and challenges of each aspect of hand and grasp control research and highlights the gaps and limitations of the current approaches. In the last part, some open questions and future directions for hand and grasp control research are suggested by emphasizing the need for a neuroscientific approach that can bridge the gap between the brain and the hand.

show abstract

Unsupervised domain adaptation by incremental learning for concept drifting data streams

Moradi,

Rahmanimanesh,

Shahzadi

2024

Int. J. Mach. Learn. & Cyber.

View full text Add to dashboard Cite

Overcoming Catastrophic Forgetting in Continual Learning by Exploring Eigenvalues of Hessian Matrix

Cited by 6 publications

References 28 publications

CL-BPUWM: continuous learning with Bayesian parameter updating and weight memory

CL-BPUWM: continuous learning with Bayesian parameter updating and weight memory

A Perspective on Prosthetic Hands Control: From the Brain to the Hand

Unsupervised domain adaptation by incremental learning for concept drifting data streams

Contact Info

Product

Resources

About