Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ComS2T: A complementary spatiotemporal learning system for data-adaptive model evolution

Zhengyang Zhou, Qihe Huang, Binwu Wang, Jianpeng Hou, Kuo Yang, Yuxuan Liang, Yang Wang

TL;DR

This work introduces a prompt-based complementary spatiotemporal learning termed ComS2T, to empower the evolution of models for data adaptation and validate the efficacy of ComS2T in adapting to various spatiotemporal out-of-distribution scenarios while maintaining efficient inference capabilities.

Abstract

Spatiotemporal (ST) learning has become a crucial technique to enable smart cities and sustainable urban development. Current ST learning models capture the heterogeneity via various spatial convolution and temporal evolution blocks. However, rapid urbanization leads to fluctuating distributions in urban data and city structures over short periods, resulting in existing methods suffering generalization and data adaptation issues. Despite efforts, existing methods fail to deal with newly arrived observations and those methods with generalization capacity are limited in repeated training. Motivated by complementary learning in neuroscience, we introduce a prompt-based complementary spatiotemporal learning termed ComS2T, to empower the evolution of models for data adaptation. ComS2T partitions the neural architecture into a stable neocortex for consolidating historical memory and a dynamic hippocampus for new knowledge update. We first disentangle two disjoint structures into stable and dynamic weights, and then train spatial and temporal prompts by characterizing distribution of main observations to enable prompts adaptive to new data. This data-adaptive prompt mechanism, combined with a two-stage training process, facilitates fine-tuning of the neural architecture conditioned on prompts, thereby enabling efficient adaptation during testing. Extensive experiments validate the efficacy of ComS2T in adapting to various spatiotemporal out-of-distribution scenarios while maintaining efficient inference capabilities.

ComS2T: A complementary spatiotemporal learning system for data-adaptive model evolution

TL;DR

This work introduces a prompt-based complementary spatiotemporal learning termed ComS2T, to empower the evolution of models for data adaptation and validate the efficacy of ComS2T in adapting to various spatiotemporal out-of-distribution scenarios while maintaining efficient inference capabilities.

Abstract

Spatiotemporal (ST) learning has become a crucial technique to enable smart cities and sustainable urban development. Current ST learning models capture the heterogeneity via various spatial convolution and temporal evolution blocks. However, rapid urbanization leads to fluctuating distributions in urban data and city structures over short periods, resulting in existing methods suffering generalization and data adaptation issues. Despite efforts, existing methods fail to deal with newly arrived observations and those methods with generalization capacity are limited in repeated training. Motivated by complementary learning in neuroscience, we introduce a prompt-based complementary spatiotemporal learning termed ComS2T, to empower the evolution of models for data adaptation. ComS2T partitions the neural architecture into a stable neocortex for consolidating historical memory and a dynamic hippocampus for new knowledge update. We first disentangle two disjoint structures into stable and dynamic weights, and then train spatial and temporal prompts by characterizing distribution of main observations to enable prompts adaptive to new data. This data-adaptive prompt mechanism, combined with a two-stage training process, facilitates fine-tuning of the neural architecture conditioned on prompts, thereby enabling efficient adaptation during testing. Extensive experiments validate the efficacy of ComS2T in adapting to various spatiotemporal out-of-distribution scenarios while maintaining efficient inference capabilities.
Paper Structure (24 sections, 23 equations, 8 figures, 5 tables, 2 algorithms)

This paper contains 24 sections, 23 equations, 8 figures, 5 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Problem formulation and basic structures
  4. Theoretical analysis for complementary spatiotemporal learning
  5. Fundamental assumptions
  6. Failure on traditional learning
  7. Superiority of complementary learning
  8. Methodology
  9. Framework overview
  10. Efficient neural disentanglement
  11. Self-supervised spatial-temporal prompt learning
  12. Progressive spatiotemporal learning
  13. Experiment
  14. Dataset description
  15. Learning with spatial-temporal OOD settings
  16. ...and 9 more sections

Figures (8)

  • Figure 1: Examples of urban dynamics along city evolution
  • Figure 2: Framework overview of ComS2T, consisting of four major components. The right top Complementary neural disentanglement is the detailed illustration of the first component for decoupling stable and dynamic weight spaces in CLS.
  • Figure 3: Experimental settings of four OOD scenarios
  • Figure 4: Hyperparameter study under both spatial and temporal shifts
  • Figure 5: Visualization of spatial-temporal prompts under different contexts.
  • ...and 3 more figures