Test-Time Adaptation with Binary Feedback

Jinwoo Shin; Junsu Kim; Sorn Chottananurak; Sung-Ju Lee; Taeckyung Lee; Taesik Gong

REVIEW 5 major objections 6 minor 22 references

Test-Time Adaptation with Binary Feedback

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Binary correct/incorrect feedback on a few uncertain samples, paired with self-adaptation on confident ones, can guide test-time adaptation past severe distribution shift and even beat full-class active labeling at equal labeling cost.

desk verdict A genuinely new TTA-with-binary-feedback setting and a plausible method, but the headline accuracy number is undefined, there are internal inconsistencies, and the agreement-based self-training path is under-validated on severe shifts. read the letter →

arxiv 2505.18514 v1 pith:QM5MHGVW submitted 2025-05-24 cs.LG cs.AI

Taeckyung Lee , Sorn Chottananurak , Junsu Kim , Jinwoo Shin , Taesik Gong , Sung-Ju Lee This is my paper · ORCID

classification cs.LGcs.AI

keywords test-timeadaptationbinaryfeedbackreinforcementlearningMC-dropoutactivedomainshiftself-adaptation

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Test-time adaptation usually tunes a pretrained model on unlabeled test data, which can fail under severe shifts. This paper proposes a cheaper alternative: ask a human only whether the model's prediction is correct or incorrect, a single bit, on a handful of uncertain samples per batch. The paper's method, BiTTA, combines this binary feedback with self-training on confidently predicted unlabeled samples, both optimized through a reinforcement-learning objective. On corruption and domain-shift benchmarks, BiTTA reports an average 13.3 percentage-point accuracy gain over state-of-the-art baselines, and it outperforms an active-test-time-adaptation method that receives full class labels under equal labeling cost. If it holds, the result means a thumb up or down on a few predictions can substitute for expensive full labels during deployment.

What carries the argument

The load-bearing object is the MC-dropout policy estimate $\pi_\theta(y|x)$, the mean softmax over $N$ stochastic forward passes, combined with REINFORCE policy gradients. MC-dropout does double duty: it selects the $k$ least-confident samples for binary feedback (BFA) and identifies agreement samples for self-adaptation (ABA). The combined loss is $L_{\text{BiTTA}} = \alpha L_{\text{BFA}} + \beta L_{\text{ABA}}$ with $L_{\text{BFA}}$ minimizing cross-entropy on correct-memory samples and maximizing it on incorrect-memory samples, and $L_{\text{ABA}}$ minimizing cross-entropy only on agreement samples. The mechanism is what lets one bit of feedback generalize: the oracle bit is sparse but direct, and the agreement path supplies the dense, unlabeled self-supervision.

What would settle it

Run BiTTA with BFA disabled (ABA-only) on a corruption type or severity not in the paper's suite and measure whether agreement-sample accuracy stays high. If on some shift the agreement samples' accuracy falls near chance while ABA-only accuracy drops below BFA-only accuracy, the self-adaptation path is amplifying noise rather than knowledge. Concretely, report the accuracy of the agreement set $S_{\text{ABA}}$ on ImageNet-C or a new severity-5 corruption and the delta between BiTTA and its BFA-only ablation.

Watch

Extended reading notes

Core claim

The central claim is that binary correct/incorrect feedback, used sparingly, is enough to guide test-time adaptation past severe distribution shift, and that it can beat full-class active labeling once labeling cost is accounted for. BiTTA formulates adaptation as policy-gradient reinforcement learning: the model's MC-dropout softmax is the policy, the predicted label is the action, and rewards come from two complementary paths. Uncertain samples (lowest MC-dropout confidence in the predicted class) receive ±1 rewards from the oracle and are used to push the model toward correct predictions and away from incorrect ones; confident samples, identified by agreement between deterministic and MC-dropout predictions, receive a +1 self-reward for maintaining their current prediction. On CIFAR10-C the method reaches 87.20% accuracy with 3 labels per 64-sample batch, versus 83.78% for the best binary-feedback-augmented TTA baseline and 83.13% for full-class SimATTA under equal sample count, and it also outperforms SimATTA under equal labeling cost.

Load-bearing premise

The method assumes that a sample whose deterministic prediction matches its MC-dropout prediction is confidently correct enough to serve as a self-training target; if that agreement no longer tracks accuracy on an unseen shift, the ABA path could reinforce the model's errors.

Editorial extensions

If this is right

If the central claim holds, a deployment pipeline can request only yes/no feedback from operators or users, cutting annotation cost by a factor of roughly $\log(\text{num\_classes})$ per label compared with full-class labeling.
BiTTA's dual-path objective means the method does not have to choose between active labeling and unlabeled self-training; both share one policy-gradient update.
The agreement-based selection criterion is dynamic and threshold-free, so it can track confidence drift across a continual stream without re-tuning a confidence threshold.
Because MC-dropout provides the uncertainty estimate with no architectural change, the approach transfers to standard pretrained backbones.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The binary-feedback setting could be extended to non-classification tasks such as segmentation or detection, where correct/incorrect is defined per prediction; the RL formulation only requires a scalar reward.
The equal-labeling-cost comparison suggests the binding constraint in active TTA may be the number of human interactions rather than the information content per label, so methods that use cheap labels to query more samples could dominate regardless of label richness.
A testable extension is to replace the oracle with a learned reward model calibrated on a handful of binary labels and measure how BiTTA's margin shrinks as reward noise increases.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The setting is the real contribution here: test-time adaptation with a few correct/incorrect feedback labels is a practical middle ground between full-class active TTA and fully unsupervised TTA, and I don't think anyone has framed it quite this way before. BiTTA's dual-path structure (BFA on uncertain samples, ABA on confident ones) is a reasonable design, and the experimental coverage is broad: three corruption benchmarks, PACS, extra datasets, ablations, sensitivity analyses, and released code. The result that binary feedback can beat full-class SimATTA under equal labeling cost is notable if it holds up.

That said, the paper as written has several soft spots. The abstract's "13.3%p accuracy improvement" is never defined and doesn't match the tables: against the best binary-feedback baseline the margins are roughly 3.4 pp on CIFAR10-C, 1.8 pp on CIFAR100-C, 1.0 pp on Tiny-ImageNet-C, and 11.2 pp on PACS, averaging around 4–5 pp, not 13.3. The authors also state alpha=beta=1 in Section 3.3, alpha=2/beta=1 in Section 4, and alpha=1 in Figure 10's caption; one of these is wrong. The REINFORCE framing is loose: Eq. 3 assumes actions sampled from pi, but the algorithm uses the deterministic argmax and then optimizes cross-entropy on MC-dropout probabilities. The actual loss is pseudo-labeling with a confidence gate, which is fine, but the RL story oversells it.

The stress-test concern about ABA is legitimate. Agreement between the deterministic and MC-dropout predictions does not guarantee correctness on severe shifts; Figure 4b only validates this on CIFAR10-C, where source accuracy is already decent. On Tiny-ImageNet-C the margin over the best baseline is only about one point, and if ABA reinforces confidently wrong targets there, the overall claim of consistent SOTA would not survive. The paper does not report agreement-sample accuracy on the harder benchmarks, so this is an unaddressed risk rather than a demonstrated failure. Missing error bars in Table 1 are a minor issue since Table 12 reports them.

Overall, the central finding—that a few binary labels can improve TTA over no labels and over full-class active TTA—is plausible and worth taking seriously. The paper needs revision to fix the numerical inconsistencies, tighten the RL framing, and add per-dataset evidence on ABA's reliability under severe shift. I would send it to peer review; it deserves referee time, but as a conditional accept, not a clean one.

Referee Report

5 major / 6 minor

Summary. This paper introduces a test-time adaptation setting in which an oracle provides binary correct/incorrect feedback on a few model predictions per batch. The authors propose BiTTA, a dual-path method combining Binary Feedback-guided Adaptation (BFA) on uncertain samples with Agreement-Based self-Adaptation (ABA) on confident samples, both expressed through a REINFORCE-style policy-gradient objective with Monte Carlo dropout. Experiments on corruption benchmarks (CIFAR10-C, CIFAR100-C, Tiny-ImageNet-C) and domain shifts (PACS) claim that BiTTA outperforms adapted TTA baselines and full-class active TTA (SimATTA) under equal labeling budgets, with the abstract reporting a 13.3 percentage-point average improvement.

Significance. If the central claims hold, the paper makes a useful contribution by proposing a cheaper supervision interface (binary correctness feedback) for test-time adaptation and showing that it can outperform methods that use full-class labels. The manuscript is strong in its breadth of experiments—multiple corruption severities, domain generalization, batch-size-1 and non-iid streams, additional architectures, and a public code release—and the ablation isolating BFA and ABA clarifies the contribution of each path. However, the significance is tempered by (i) the undefined and internally inconsistent headline improvement, (ii) a mismatch between the stated policy-gradient derivation and the implemented loss, and (iii) an unvalidated assumption in the ABA selection rule that is load-bearing for the accuracy claims on the most severe shifts.

major comments (5)

[Abstract / Section 4] The claimed "13.3%p accuracy improvement over state-of-the-art baselines" is not defined and is inconsistent with the detailed tables: in Table 1, BiTTA's margin over the strongest binary-feedback baseline is +3.42 pp on CIFAR10-C, +1.79 pp on CIFAR100-C, and +1.03 pp on Tiny-ImageNet-C, while in Table 2 the margin is +12.94 pp on PACS domain-wise and +2.97 pp on PACS mixed; no averaging protocol in the text produces 13.3 pp. Please specify the baseline, the set of benchmarks, and the aggregation rule, and update the abstract accordingly.
[Section 3, Eq. (3), Algorithm 1, Eq. (12)] The REINFORCE estimator in Eq. (3) is derived for actions sampled from π_θ, but Algorithm 1 computes the gradient for the deterministic prediction y* = argmax f_θ(x); this is a biased estimator of ∇J and the connection between Eq. (3) and the implemented cross-entropy loss in Eq. (12) is not established. Please either sample actions from the MC-dropout policy during training or reframe the objective as a weighted self-training loss and remove the policy-gradient claim.
[Section 3.2, Eq. (8)-(9), Section 4 ablation] The ABA selection rule assumes that agreement between the deterministic prediction and the MC-dropout argmax implies the pseudo-label is correct; the validation for this is only Figure 4b on CIFAR10-C, where source accuracy is already 57%. On CIFAR100-C and Tiny-ImageNet-C, source accuracy is 33% and 21%, where a model can be consistently wrong across dropout masks, so agreement may select confidently wrong targets; since ABA-only accounts for 82.64% of BiTTA's 87.20% on CIFAR10-C, a failure of this assumption would substantially erode the headline accuracy. Please report the accuracy of S_ABA versus its complement per corruption type for CIFAR100-C and Tiny-ImageNet-C, and show the online accuracy trajectory of ABA-selected samples to demonstrate the assumption holds under severe shifts.
[Section 3.3 and Section 4] The main balancing hyperparameters are stated inconsistently: Section 3.3 says "We set α=β=1 in all experiments", Section 4 says "α=2 and β=1", and the Figure 10 caption says α is set to 1. This ambiguity directly affects the reported objective and the sensitivity analysis; please correct the statements and ensure the figure matches the main configuration.
[Section 4, Figure 5, Appendix D.1] The "equal labeling cost" comparison that supports the claim "up to 32% higher accuracy" is not fully specified: Figure 5 displays "+9%, +30%, +32%" without defining whether these are relative or percentage-point improvements, and the number of feedback samples used by each method under the Shannon-cost rule is described only in the appendix; clarify the exact budgets (e.g., 3 binary vs 1 full-class sample for CIFAR-10) and state the error bars and significance of the differences.

minor comments (6)

[Figures 1, 5, and Appendix B] Figures 1, 5, and several Appendix B figures contain garbled unicode characters in their captions (e.g., "/uni00000028/uni00000057/..." in Figure 5), which appear to be a rendering artifact; the figures and captions should be regenerated.
[Figure 2] Figure 2's x-axis labels are concatenated as "TENT*EATA*CoTTA*RoTTA*SoTTA*BiTTA"; please use separate labels and a legend.
[Eq. (10)] In Eq. (10), the ABA term β E_{x∈S_ABA}[R_ABA(x,y)] is constant (=β) since R_ABA=1 for all x∈S_ABA by definition; clarify that the actual update uses the gradient E[∇ log π] rather than the literal objective.
[Tables 1 and others] Table 1 and several other tables report only averages over three seeds without standard deviations; given that Table 2 reports standard deviations and some baselines (e.g., SimATTA* on PACS) have very high variance, error bars should be reported consistently.
[Section 4, baseline adaptations] The description of the binary-feedback baseline modification says "an equal number of random binary-feedback data" but does not state the selection rule (random vs uncertainty) or the exact loss weighting for each adapted baseline; please specify the full protocol for reproducibility.
[Appendix D.1] The statement that BiTTA uses "three binary feedback samples within each 64-sample test batch, accounting for less than 5%" is correct, but the later claim of "log(num_class) times more feedback samples" under equal labeling cost should be made more explicit with the actual numbers per dataset.

Circularity Check

0 steps flagged · score 1.0 of 10

No circular derivation: BiTTA's accuracy gains are measured on held-out public corruption/domain-shift benchmarks against external baselines; self-citations (SoTTA, AETTA) supply pretrained weights or dropout placement, not the target result.

full rationale

The paper's central claims are empirical and are evaluated on public benchmark test streams (CIFAR10-C, CIFAR100-C, Tiny-ImageNet-C, PACS) against external baselines. The dual-path objective in Eq. (10)-(12) is a standard policy-gradient/self-training formulation: BFA uses oracle binary feedback as reward, and ABA uses the model's own agreement-based pseudo-labels. ABA's selection rule (Eq. 8) and reward (Eq. 9) do define the target as the model's own prediction, but this is self-training, not a fitted parameter being relabeled as a prediction; its benefit is verified by held-out accuracy comparisons and ablations, not derived from the definition. The equal-labeling-cost comparison with SimATTA is an explicit information-theoretic cost model (log(num_class) bits vs. 1 bit), not a circular construction. The only self-citations that appear load-adjacent are SoTTA (Gong et al., 2023b) and AETTA (Lee et al., 2024b): SoTTA provides the pretrained source weights used uniformly for all methods, and AETTA is cited for dropout placement. These supply experimental infrastructure and design details rather than the central accuracy claim, and the benchmarks themselves are external and reproducible. No step of the derivation reduces by definition to its own inputs, and no 'prediction' is obtained by fitting a parameter to the quantity it is then said to predict. The weakest point is an unvalidated assumption about ABA's agreement criterion on severe shifts, but that is a correctness/robustness risk, not circularity.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The paper's central claim rests on standard ML machinery (REINFORCE, MC-dropout, complementary loss) and on two domain assumptions: MC-dropout confidence is calibrated enough for sample selection, and agreement between deterministic and dropout predictions identifies reliable pseudo-labels. The equal-cost comparison against full-class active TTA adds a bit-cost model that is favorable to BiTTA. No new physical entities are introduced.

free parameters (7)

alpha (BFA weight) = 2 (Section 4); 1 (Section 3.3)
Balances binary-feedback-guided adaptation; the paper is internally inconsistent about the value used.
beta (ABA weight) = 1
Balances agreement-based self-adaptation; Figure 10 sweeps beta with alpha fixed at 1.
k (feedback samples per batch) = 3
Number of binary-feedback queries per 64-sample batch, described as less than 5% labeling budget.
N (MC-dropout iterations) = 4
Number of stochastic forward passes for policy and uncertainty estimation; N>=2 is shown to be robust.
dropout rate = 0.3 (small-scale), 0.1 (large-scale)
Dropout rate for MC-dropout; set per dataset scale, no systematic sweep reported.
learning rate and epochs per dataset = 0.001/3 (PACS), 0.0001/3 (CIFAR10/100-C), 0.00005/5 (Tiny-ImageNet-C)
SGD optimizer hyperparameters chosen per dataset; results depend on these choices.
BN momentum = 0.3
Batch norm statistics updated on the test batch then frozen during adaptation.

assumptions (4)

standard math REINFORCE gradient estimator (Williams, 1992) provides an unbiased estimate of the expected-reward gradient (Eq. 3).
The paper builds the objective in Eq. 2 on this estimator, but the action is the deterministic argmax, not a sample from the policy, so the application is informal.
domain assumption MC-dropout mean approximates the predictive distribution and provides calibrated uncertainty under distribution shift (Eq. 4).
Used to select uncertain samples for BFA and agreement samples for ABA; ECE measurements support it on CIFAR10-C but it is assumed across datasets and architectures.
domain assumption Agreement between deterministic and MC-dropout argmax identifies samples whose model predictions are reliable enough to use as self-training targets (Eq. 8).
Load-bearing for ABA, the higher-contributing path (82.64% alone vs 87.20% combined); supporting evidence is Figure 4b on CIFAR10-C only.
domain assumption Equal labeling-cost comparison: binary feedback costs 1 bit and full-class label costs log(num_class) bits (Appendix D.1).
Used to give BiTTA log(C) times more feedback samples than SimATTA in Figure 5; if binary feedback is not actually cheaper in annotator time, the comparison is biased toward BiTTA.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Test-Time Adaptation with Binary Feedback." pith.science (2026). https://pith.science/paper/QM5MHGVW

@misc{pith2026250518514,
  author       = {Pith},
  title        = {Pith review of: Test-Time Adaptation with Binary Feedback},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QM5MHGVW}},
  note         = {Machine review of arXiv:2505.18514}
}

read the original abstract

Deep learning models perform poorly when domain shifts exist between training and test data. Test-time adaptation (TTA) is a paradigm to mitigate this issue by adapting pre-trained models using only unlabeled test samples. However, existing TTA methods can fail under severe domain shifts, while recent active TTA approaches requiring full-class labels are impractical due to high labeling costs. To address this issue, we introduce a new setting of TTA with binary feedback. This setting uses a few binary feedback inputs from annotators to indicate whether model predictions are correct, thereby significantly reducing the labeling burden of annotators. Under the setting, we propose BiTTA, a novel dual-path optimization framework that leverages reinforcement learning to balance binary feedback-guided adaptation on uncertain samples with agreement-based self-adaptation on confident predictions. Experiments show BiTTA achieves 13.3%p accuracy improvements over state-of-the-art baselines, demonstrating its effectiveness in handling severe distribution shifts with minimal labeling effort. The source code is available at https://github.com/taeckyung/BiTTA.

Figures

Figures reproduced from arXiv: 2505.18514 by the authors.

**Figure 3.** Overview of BiTTA algorithm. BiTTA implements a reinforcement learning-based dual-path optimization that [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

**Figure 4.** Analysis of confidence and accuracy during online adaptation. (a) Average sample-wise confidence over time [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

Figures from the paper (3 more)

**Figure 5.** Figure 5: Accuracy (%) with full-class feedback (SimATTA) and binary-feedback (BiTTA) and under the equal total labeling cost. GPT-4o is used as a foundational model to provide a full-class label. 0 10 20 30 Feedback error ratio (%) 60 70 80 90 Accuracy (%) BiTTA SimATTA* [PIT…

**Figure 9.** Figure 9: Accuracy (%) varying the binary-feedback sample selection strategy in CIFAR10-C. Ours leverages MC-dropout to select the most uncertain samples. Averaged over three random seeds. 0.5 1.0 1.5 2.0 Hyperparameter 50 60 70 80 90 Accuracy (%) CIFAR10-C PACS [PITH_FULL_IMA…

**Figure 11.** Figure 11: Accuracy (%) varying the labeling skip in CIFAR10-C. Averaged over three random seeds. 0 1 2 3 Number of delayed batches 70 75 80 85 90 Accuracy (%) BiTTA SimATTA* [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

22 extracted references · 19 canonical work pages

[1]

Invariant risk minimization.arXiv preprint arXiv:1907.02893,

Arjovsky, M., Bottou, L., Gulrajani, I., and Lopez-Paz, D. Invariant risk minimization.arXiv preprint arXiv:1907.02893,

arXiv 1907
[3]

could be further reduced by efficient TTA (Hong et al., 2023; Song et al.,

work page 2023
[4]

Compared to baseline methods, BiTTA shows higher memory usage due to repeated forward passes for MC- dropout and the reinforcement learning formulation. To mitigate this, we evaluate two variants: BiTTA+MECTA (applying the memory-efficient continual test-time adaptation from MECTA (Hong et al., 2023)) and BiTTA+GA (using gradient accumulation to divide a ...

work page 2023
[5]

Notation * indicates the modified algorithm to utilize binary-feedback samples

with additional scenarios. Notation * indicates the modified algorithm to utilize binary-feedback samples. B: TTA with binary feedback. Results outperforming all other baselines are highlighted inboldfonts. Averaged over three random seeds. Noise Blur Weather Digital Label Method Gau. Shot Imp. Def. Gla. Mot. Zoom Snow Fro. Fog Brit. Cont. Elas. Pix. JPEG...

work page arXiv
[6]

As shown in Figure 13, BiTTA performs consistently well for all N >1 , demonstrating robustness to this hyperparameter. Even at N= 1 , MC-dropout is still active, introducing stochasticity essential for both core components: (1) BFA relies on uncertainty to select feedback samples, and (2) ABA leverages agreement between stochastic and deterministic predi...

work page 2022
[8]

0 1 2 3 Number of skipped batches 70 75 80 85 90Accuracy (%) BiTTA SimATTA* Figure 11: Accuracy (%) varying the labeling skip in CIFAR10-C

Averaged over three random seeds. 0 1 2 3 Number of skipped batches 70 75 80 85 90Accuracy (%) BiTTA SimATTA* Figure 11: Accuracy (%) varying the labeling skip in CIFAR10-C. Aver- aged over three random seeds. 0 1 2 3 Number of delayed batches 70 75 80 85 90Accuracy (%) BiTTA SimATTA* Figure 12: Accuracy (%) varying the labeling delay (labeled samples arr...

work page 2020
[11]

Comparison with recent baselinesWe incorporated new comparisons with recent TTA methods: DeYO (Lee et al., 2024a), OWTTT (Li et al., 2023), and TAST (Jang et al., 2022)

suggest the robustness of our method over imbalanced label distribution and batch size 1 by effectively utilizing reward signals from the binary feedback and unlabeled samples. Comparison with recent baselinesWe incorporated new comparisons with recent TTA methods: DeYO (Lee et al., 2024a), OWTTT (Li et al., 2023), and TAST (Jang et al., 2022). As shown i...

work page 2023
[13]

Results on additional architectures.To further examine the applicability of BiTTA in various model architectures, we experimented with ResNet50 and ViT-Base

outperforms these methods under equal binary feedback conditions. Results on additional architectures.To further examine the applicability of BiTTA in various model architectures, we experimented with ResNet50 and ViT-Base. Table 11 shows the overall result, where BiTTA still outperformed the baselines in all corruptions, demonstrating the applicability o...

work page 2024

Show all 22 references

[14]

Notation * indicates the modified algorithm to utilize binary-feedback samples

with additional architectures. Notation * indicates the modified algorithm to utilize binary-feedback samples. B: TTA with binary feedback. Results outperforming all other baselines are highlighted inboldfonts. Averaged over three random seeds. Noise Blur Weather Digital Label...

2019
[17]

Notation * indicates the modified algorithm to utilize binary-feedback samples

and TTA with binary feedback baselines on source model, in CIFAR10-C (severity level 5). Notation * indicates the modified algorithm to utilize binary-feedback samples. E: Enhanced TTA, B: TTA with binary feedback. Results outperforming all other baselines are highlighted inbo...

1981
[18]

We assume the probability of each feedback strategy as p= 2−1 (correct/incorrect) and p=num_class −1 (select in the entire class set)

as log(p−1) where p is the probability of selecting a label. We assume the probability of each feedback strategy as p= 2−1 (correct/incorrect) and p=num_class −1 (select in the entire class set). The final labeling cost for binary feedback is 1 for binary feedback and log(num_...

2021
[20]

The implementation followed the official code.4 RoTTA.For RoTTA (Yuan et al., 2023), we utilized the Adam optimizer (Kingma & Ba,

asp th = 0.1. The implementation followed the official code.4 RoTTA.For RoTTA (Yuan et al., 2023), we utilized the Adam optimizer (Kingma & Ba,

2023
[21]

with a learning rate of LR= 0.001 and β= 0.9. We followed the original hyperparameters, including BN-statistic exponential moving average updating rate α= 0.05 , Teacher model’s exponential moving average updating rate ν= 0.001 , timeliness parameter λt = 1.0, and uncertainty ...

2015
[22]

We update batch norm statistics with momentum 0.3 on the unlabeled test batch, and freeze the statistics during adaptation, following Gui et al

in Tiny-ImageNet-C to prevent overfitting. We update batch norm statistics with momentum 0.3 on the unlabeled test batch, and freeze the statistics during adaptation, following Gui et al. (2024). We apply the dropout layer after residual blocks, following the previous work on ...

2024
[64]

The number of fine-tuning epochs was set to 150 for PACS, 150 for CIFAR-10, 150 for CIFAR-100, and 25 18 Test-Time Adaptation with Binary Feedback Table 13: Accuracy (%) and standard deviation comparisons withoriginal TTA and full-label active TTA baselinesin PACS. The domain-...

2022
[2003]

Evaluating prediction-time batch normalization for robustness under covariate shift

Nado, Z., Padhy, S., Sculley, D., D’Amour, A., Lakshmi- narayanan, B., and Snoek, J. Evaluating prediction-time batch normalization for robustness under covariate shift. arXiv preprint arXiv:2006.10963,

2006 arXiv
[2015]

with a learning rate LR= 0.001 , aligning with the guidelines outlined in the original paper and active TTA paper (Gui et al., 2024). The implementation followed the official code.1 EATA.For EATA (Niu et al., 2022), we followed the original configuration of LR= 0.001 , entropy...

2024
[2016]

from TorchVision (TorchVision maintainers and contributors, 2016). For CIFAR10-C/CIFAR100-C/Tiny-ImageNet-C, we trained the model with the source data with a learning rate of 0.1/0.1/0.001 and a momentum of 0.9, with cosine annealing learning rate scheduling for 200 epochs. Fo...

2016
[2019]

U., Teterwak, P., et al

Bashkirova, D., Hendrycks, D., Kim, D., Liao, H., Mishra, S., Rajagopalan, C., Saenko, K., Saito, K., Tayyab, B. U., Teterwak, P., et al. Visda-2021 competition: Universal domain adaptation to improve performance on out-of- distribution data. InNeurIPS 2021 Competitions and De...

2021
[2022]

Y ., Acer, U

Gong, T., Jang, S. Y ., Acer, U. G., Kawsar, F., and Min, C. Collaborative inference via dynamic compo- sition of tiny ai accelerators on mcus.arXiv preprint arXiv:2401.08637, 2023a. 10 Test-Time Adaptation with Binary Feedback Gong, T., Kim, Y ., Lee, T., Chottananurak, S., a...

arXiv
[2023]

and on-device machine learning (Liberis & Lane, 2023; Rusci et al., 2020; Gong et al., 2023a). Finally, although our algorithm robustly outperformed with feedback errors (Figure 6), designing a method for specifically handling noisy or incorrect feedback remains an area for fu...

2023
[2024]

Con- former: Convolution-augmented transformer for speech recognition.arXiv preprint arXiv:2005.08100,

Gulati, A., Qin, J., Chiu, C.-C., Parmar, N., Zhang, Y ., Yu, J., Han, W., Wang, S., Zhang, Z., Wu, Y ., et al. Con- former: Convolution-augmented transformer for speech recognition.arXiv preprint arXiv:2005.08100,

2005 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Invariant risk minimization.arXiv preprint arXiv:1907.02893,

Arjovsky, M., Bottou, L., Gulrajani, I., and Lopez-Paz, D. Invariant risk minimization.arXiv preprint arXiv:1907.02893,

arXiv 1907

[2] [3]

could be further reduced by efficient TTA (Hong et al., 2023; Song et al.,

work page 2023

[3] [4]

Compared to baseline methods, BiTTA shows higher memory usage due to repeated forward passes for MC- dropout and the reinforcement learning formulation. To mitigate this, we evaluate two variants: BiTTA+MECTA (applying the memory-efficient continual test-time adaptation from MECTA (Hong et al., 2023)) and BiTTA+GA (using gradient accumulation to divide a ...

work page 2023

[4] [5]

Notation * indicates the modified algorithm to utilize binary-feedback samples

with additional scenarios. Notation * indicates the modified algorithm to utilize binary-feedback samples. B: TTA with binary feedback. Results outperforming all other baselines are highlighted inboldfonts. Averaged over three random seeds. Noise Blur Weather Digital Label Method Gau. Shot Imp. Def. Gla. Mot. Zoom Snow Fro. Fog Brit. Cont. Elas. Pix. JPEG...

work page arXiv

[5] [6]

As shown in Figure 13, BiTTA performs consistently well for all N >1 , demonstrating robustness to this hyperparameter. Even at N= 1 , MC-dropout is still active, introducing stochasticity essential for both core components: (1) BFA relies on uncertainty to select feedback samples, and (2) ABA leverages agreement between stochastic and deterministic predi...

work page 2022

[6] [8]

0 1 2 3 Number of skipped batches 70 75 80 85 90Accuracy (%) BiTTA SimATTA* Figure 11: Accuracy (%) varying the labeling skip in CIFAR10-C

Averaged over three random seeds. 0 1 2 3 Number of skipped batches 70 75 80 85 90Accuracy (%) BiTTA SimATTA* Figure 11: Accuracy (%) varying the labeling skip in CIFAR10-C. Aver- aged over three random seeds. 0 1 2 3 Number of delayed batches 70 75 80 85 90Accuracy (%) BiTTA SimATTA* Figure 12: Accuracy (%) varying the labeling delay (labeled samples arr...

work page 2020

[7] [11]

Comparison with recent baselinesWe incorporated new comparisons with recent TTA methods: DeYO (Lee et al., 2024a), OWTTT (Li et al., 2023), and TAST (Jang et al., 2022)

suggest the robustness of our method over imbalanced label distribution and batch size 1 by effectively utilizing reward signals from the binary feedback and unlabeled samples. Comparison with recent baselinesWe incorporated new comparisons with recent TTA methods: DeYO (Lee et al., 2024a), OWTTT (Li et al., 2023), and TAST (Jang et al., 2022). As shown i...

work page 2023

[8] [13]

Results on additional architectures.To further examine the applicability of BiTTA in various model architectures, we experimented with ResNet50 and ViT-Base

outperforms these methods under equal binary feedback conditions. Results on additional architectures.To further examine the applicability of BiTTA in various model architectures, we experimented with ResNet50 and ViT-Base. Table 11 shows the overall result, where BiTTA still outperformed the baselines in all corruptions, demonstrating the applicability o...

work page 2024