MemFLoRA: Memory-Floor LoRA for CNN Adaptation at the Edge

Key points

• On-device learning is necessary when the model encounters user-,sensor-, or environment-specific shifts after deployment.1
• Although parameter-efficient fine-tuning (PEFT) methods, particularly Low-Rank Adaptation (LoRA) variants, enable efficient adaptation at the edge, the limiting resource for Convolutional Neural Network (CNN) adaptation is often not the number of trainable parameters but the activation state that must be retained until the backward pass.1
• Instead of merely reducing trainable weights, we define an activation-memory-floor criterion: trainable backward computations must not depend on full-width layer inputs.1
• Evaluated on three Human Activity Recognition (HAR) datasets and two CNN backbones under subject, body-location, and sensor-placement shifts, MemFLoRA reduces saved-activation memory by 98.5-98.7% and peak training-state memory by 94.9-97.3% relative to full fine-tuning, while matching or exceeding CNN PEFT baselines.1

Sources (1)

• [1]MemFLoRA: Memory-Floor LoRA for CNN Adaptation at the Edge
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 04:51 PM

  This paper introduces Memory-Floor LoRA (MemFLoRA), a low-rank CNN adapter built around a memory-first design principle rather than a direct application of transformer-oriented LoRA.

  On-device learning is necessary when the model encounters user-,sensor-, or environment-specific shifts after deployment.

Extractive summary: sentences quoted from the sources.