CLIFF: Continual Learning for Incremental Flake Features in 2D Material Identification

2025-08-24

AI Summary

Accurate layer-number classification of quantum flakes in optical microscopy images of 2D materials is hindered by substantial inter-material appearance variations and severe catastrophic forgetting during continual learning. Method: This work introduces continual learning to this domain for the first time, proposing a feature modulation mechanism based on a prompt pool and cosine-similarity gating. It employs a frozen backbone, material-specific prompt vectors, prompt-embedded layers, and an incremental delta-head architecture, integrated with memory replay and knowledge distillation. Contribution/Results: The method substantially mitigates catastrophic forgetting while maintaining high classification accuracy across multi-material incremental scenarios. It reduces forgetting rates significantly compared to conventional fine-tuning and prompt-based baselines. This establishes a scalable, low-forgetting paradigm for automated, intelligent characterization of 2D materials.

Technology Category

• Machine Learning: Quantum Machine Learning
• Search and Optimization: Learning to Search
• Computer Vision: Representation Learning for Vision

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Identifying quantum flakes is crucial for scalable quantum hardware; however, automated layer classification from optical microscopy remains challenging due to substantial appearance shifts across different materials. In this paper, we propose a new Continual-Learning Framework for Flake Layer Classification (CLIFF). To our knowledge, this is the first systematic study of continual learning in the domain of two-dimensional (2D) materials. Our method enables the model to differentiate between materials and their physical and optical properties by freezing a backbone and base head trained on a reference material. For each new material, it learns a material-specific prompt, embedding, and a delta head. A prompt pool and a cosine-similarity gate modulate features and compute material-specific corrections. Additionally, we incorporate memory replay with knowledge distillation. CLIFF achieves competitive accuracy with significantly lower forgetting than naive fine-tuning and a prompt-based baseline.

Problem

• Automated layer classification from optical microscopy of 2D materials
• Addressing substantial appearance shifts across different quantum materials
• Enabling continual learning for incremental flake feature identification

Innovation

• Continual learning framework for material classification
• Material-specific prompts and delta heads
• Memory replay with knowledge distillation