Pith. sign in

REVIEW 5 cited by

Test-Time Model Adaptation with Only Forward Passes

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2404.01650 v2 pith:MXLHYVBT submitted 2024-04-02 cs.LG

Shuaicheng Niu , Chunyan Miao , Guohao Chen , Pengcheng Wu , Peilin Zhao This is my paper · ORCID
classification cs.LG
keywords modeladaptationtest-timebackpropagationoftenquantizedsamplesstrategy
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Test-time adaptation has proven effective in adapting a given trained model to unseen test samples with potential distribution shifts. However, in real-world scenarios, models are usually deployed on resource-limited devices, e.g., FPGAs, and are often quantized and hard-coded with non-modifiable parameters for acceleration. In light of this, existing methods are often infeasible since they heavily depend on computation-intensive backpropagation for model updating that may be not supported. To address this, we propose a test-time Forward-Optimization Adaptation (FOA) method. In FOA, we seek to solely learn a newly added prompt (as model's input) via a derivative-free covariance matrix adaptation evolution strategy. To make this strategy work stably under our online unsupervised setting, we devise a novel fitness function by measuring test-training statistic discrepancy and model prediction entropy. Moreover, we design an activation shifting scheme that directly tunes the model activations for shifted test samples, making them align with the source training domain, thereby further enhancing adaptation performance. Without using any backpropagation and altering model weights, FOA runs on quantized 8-bit ViT outperforms gradient-based TENT on full-precision 32-bit ViT, while achieving an up to 24-fold memory reduction on ImageNet-C.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 5 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Audio-Visual Continual Test-Time Adaptation without Forgetting

    cs.LG 2026-02 conditional novelty 6.0 of 10

    By adapting only the fusion layer and retrieving past good parameter states via raw input statistics, AV-CTTA outperforms existing audio-visual continual test-time adaptation methods and forgets far less.

  2. From LLMs to Edge: Parameter-Efficient Fine-Tuning on Edge Devices

    cs.LG 2025-07 conditional novelty 6.0 of 10

    On depthwise-separable CNNs, PEFT memory savings drop to about half of LLM levels, though LoRA and DoRA still cut update FLOPs by up to 95%.

  3. When Small Guides Large: Cross-Model Co-Learning for Test-Time Adaptation

    cs.CV 2025-06 conditional novelty 6.0 of 10

    COCA shows that bidirectional cross-model co-learning during test-time adaptation lets even small models significantly improve large models' robustness to distribution shift.

  4. PAID: Pairwise Angular-Invariant Decomposition for Continual Test-Time Adaptation

    cs.CV 2025-06 reject novelty 5.0 of 10

    PAID proposes Householder-based orthogonal weight updates for continual test-time adaptation, claiming that preserving pairwise angular structure of pretrained weights is a useful prior, but the math and validation fo...

  5. EdgeMLBalancer: A Self-Adaptive Approach for Dynamic Model Switching on Resource-Constrained Edge Devices

    cs.SE 2025-02 reject novelty 3.0 of 10

    EdgeMLBalancer uses epsilon-greedy switching among four edge object-detection models, driven by CPU usage and confidence, and reports better accuracy and fairness than two baselines in a single 30-minute smartphone trial.

Pith tools