FearNet: Brain-Inspired Model for Incremental Learning

TL;DR

This work tackles incremental class learning and catastrophic forgetting in neural networks. It introduces FearNet, a brain-inspired architecture that combines a hippocampal-like short-term memory (HC), a medial prefrontal cortex-like long-term store (mPFC), and a basolateral amygdala-like selector (BLA), augmented by sleep-based pseudorehearsal to consolidate memories without storing past data. FearNet achieves state-of-the-art performance on CIFAR-100, CUB-200, and AudioSet while maintaining a compact memory footprint, demonstrating effective memory consolidation and recall across modalities. The study highlights the practicality of dual-memory systems with generative replay for scalable, on-device incremental learning.

Abstract

Incremental class learning involves sequentially learning classes in bursts of examples from the same class. This violates the assumptions that underlie methods for training standard deep neural networks, and will cause them to suffer from catastrophic forgetting. Arguably, the best method for incremental class learning is iCaRL, but it requires storing training examples for each class, making it challenging to scale. Here, we propose FearNet for incremental class learning. FearNet is a generative model that does not store previous examples, making it memory efficient. FearNet uses a brain-inspired dual-memory system in which new memories are consolidated from a network for recent memories inspired by the mammalian hippocampal complex to a network for long-term storage inspired by medial prefrontal cortex. Memory consolidation is inspired by mechanisms that occur during sleep. FearNet also uses a module inspired by the basolateral amygdala for determining which memory system to use for recall. FearNet achieves state-of-the-art performance at incremental class learning on image (CIFAR-100, CUB-200) and audio classification (AudioSet) benchmarks.

Figures (7)

• Figure 1: FearNet consists of three brain-inspired modules based on 1) mPFC (long-term storage), 2) HC (recent storage), and 3) BLA for determining whether to use mPFC or HC for recall.
• Figure 2: iCaRL's performance depends heavily on the number of exemplars per class (EPC) that it stores. Reducing EPC from 20 (blue) to 1 (red) severely impairs its ability to recall older information.
• Figure 3: The mPFC and BLA sub-systems in FearNet. mPFC is responsible for the long-term storage of remote memories. BLA is used during prediction time to determine if the memory should be recalled from short- or long-term memory.
• Figure 4: Mean-class test accuracy of all classes seen so far.
• Figure 5: FearNet performance as the sleep frequency decreases.