Memory Efficient Neural Processes via Constant Memory Attention Block

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Feng, Leo, Tung, Frederick, Hajimirsadeghi, Hossein, Bengio, Yoshua, Ahmed, Mohamed Osama
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914812199436288
author Feng, Leo
Tung, Frederick
Hajimirsadeghi, Hossein
Bengio, Yoshua
Ahmed, Mohamed Osama
author_facet Feng, Leo
Tung, Frederick
Hajimirsadeghi, Hossein
Bengio, Yoshua
Ahmed, Mohamed Osama
contents Neural Processes (NPs) are popular meta-learning methods for efficiently modelling predictive uncertainty. Recent state-of-the-art methods, however, leverage expensive attention mechanisms, limiting their applications, particularly in low-resource settings. In this work, we propose Constant Memory Attentive Neural Processes (CMANPs), an NP variant that only requires constant memory. To do so, we first propose an efficient update operation for Cross Attention. Leveraging the update operation, we propose Constant Memory Attention Block (CMAB), a novel attention block that (i) is permutation invariant, (ii) computes its output in constant memory, and (iii) performs constant computation updates. Finally, building on CMAB, we detail Constant Memory Attentive Neural Processes. Empirically, we show CMANPs achieve state-of-the-art results on popular NP benchmarks while being significantly more memory efficient than prior methods.
format Preprint
id arxiv_https___arxiv_org_abs_2305_14567
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Memory Efficient Neural Processes via Constant Memory Attention Block
Feng, Leo
Tung, Frederick
Hajimirsadeghi, Hossein
Bengio, Yoshua
Ahmed, Mohamed Osama
Machine Learning
Computer Vision and Pattern Recognition
Neural Processes (NPs) are popular meta-learning methods for efficiently modelling predictive uncertainty. Recent state-of-the-art methods, however, leverage expensive attention mechanisms, limiting their applications, particularly in low-resource settings. In this work, we propose Constant Memory Attentive Neural Processes (CMANPs), an NP variant that only requires constant memory. To do so, we first propose an efficient update operation for Cross Attention. Leveraging the update operation, we propose Constant Memory Attention Block (CMAB), a novel attention block that (i) is permutation invariant, (ii) computes its output in constant memory, and (iii) performs constant computation updates. Finally, building on CMAB, we detail Constant Memory Attentive Neural Processes. Empirically, we show CMANPs achieve state-of-the-art results on popular NP benchmarks while being significantly more memory efficient than prior methods.
title Memory Efficient Neural Processes via Constant Memory Attention Block
topic Machine Learning
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2305.14567