Fast Spatial Memory with Elastic Test-Time Training

2026-04-08

AI Summary

This work addresses the challenges of catastrophic forgetting, overfitting, and memory bottlenecks in test-time training (LaCT) for long-sequence 3D/4D reconstruction, which arise from fully plastic parameter updates. To mitigate these issues, the authors propose an elastic test-time training mechanism that integrates Fisher information-based elastic regularization and exponential moving average anchor states, thereby enhancing model stability while preserving rapid adaptation capabilities. Furthermore, they introduce a Fast Spatial Memory architecture that overcomes single-chunk processing limitations, alleviates activation memory pressure, and prevents shortcut learning through camera interpolation. By adapting the elastic weight consolidation principle—previously used in continual learning—to the test-time training paradigm, this approach enables efficient, high-quality reconstruction of arbitrarily long sequences, significantly improving generalization and reducing memory consumption.

Abstract

Large Chunk Test-Time Training (LaCT) has shown strong performance on long-context 3D reconstruction, but its fully plastic inference-time updates remain vulnerable to catastrophic forgetting and overfitting. As a result, LaCT is typically instantiated with a single large chunk spanning the full input sequence, falling short of the broader goal of handling arbitrarily long sequences in a single pass. We propose Elastic Test-Time Training inspired by elastic weight consolidation, that stabilizes LaCT fast-weight updates with a Fisher-weighted elastic prior around a maintained anchor state. The anchor evolves as an exponential moving average of past fast weights to balance stability and plasticity. Based on this updated architecture, we introduce Fast Spatial Memory (FSM), an efficient and scalable model for 4D reconstruction that learns spatiotemporal representations from long observation sequences and renders novel view-time combinations. We pre-trained FSM on large-scale curated 3D/4D data to capture the dynamics and semantics of complex spatial environments. Extensive experiments show that FSM supports fast adaptation over long sequences and delivers high-quality 3D/4D reconstruction with smaller chunks and mitigating the camera-interpolation shortcut. Overall, we hope to advance LaCT beyond the bounded single-chunk setting toward robust multi-chunk adaptation, a necessary step for generalization to genuinely longer sequences, while substantially alleviating the activation-memory bottleneck.

Problem

• catastrophic forgetting
• overfitting
• long-context reconstruction
• test-time training
• memory bottleneck

Innovation

• Elastic Test-Time Training
• Fast Spatial Memory
• LaCT
• 4D Reconstruction
• Catastrophic Forgetting