- Rating
- 3
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
The paper investigates the “bypass” problem in implicit visual reasoning, where models may answer through original visual tokens or multimodal KV caches without relying on their latent states. It introduces CVRR, which frames latent-state reliance as a causal path-design problem and uses image-conditioned initialization, learned recurrent updates, and repeated visual rereading to make the recurrent latent state necessary for prediction while preserving pretrained visual competence. Experiments on V*, MMVP, BLINK, and MME-RealWorld-Lite, together with no-bypass baseline comparisons, component ablations, and causal interventions, show that CVRR retains relatively strong performance under this constraint and that its predictions depend on recurrent states continually refined by visual evidence.
Critical strengths
-
Novel problem perspective and sound methodological design. The paper distinguishes between the informativeness of hidden states and pathway necessity. By eliminating the visual bypass via CVRR, it preserves pretrained visual capabilities while making the recurrent hidden state a necessary channel for prediction.
-
Comprehensive validation framework. Beyond multi-benchmark comparisons, the paper employs component ablation, state replacement, same-question cross-image swapping, and visual region intervention, adequately demonstrating that recurrent states and continuous visual re-reading indeed influence predictions.
Critical weaknesses
-
While the paper demonstrates causal utilization and visual refinement, it has not yet fully substantiated multi-step implicit reasoning in the stronger sense; moreover, some necessity conclusions are directly guaranteed by the architecture itself. This limits the paper’s claim from demonstrating multi-step latent reasoning to demonstrating causal latent-state utilization and iterative visual refinement.
-
Although the no-bypass baselines were uniformly retrained, issues remain regarding interface adaptation, fairness in hyperparameter tuning, and incomplete comparison coverage. Therefore, the magnitude of CVRR’s advantage over prior latent reasoning methods remains uncertain.
Questions for authors
-
Current experiments show that CVRR's predictions depend on recurrent states and that these states continuously read visual information, but is this sufficient to distinguish "genuine multi-step implicit reasoning" from "repeated feature refinement or information compression over the same visual representation"? Could the authors provide more direct controls, for example, a parameter- and compute-matched non-recurrent bottleneck, or interventions that selectively disable visual re-reading at individual recurrent steps?
-
Could the authors report the method-specific hyperparameter search, training budget, and multi-seed results for the no-bypass baselines, and clarify how much of their performance degradation may arise from interface mismatch rather than bypass dependence?
4 comments
Response to Reviewer UbSm (Part 2/3)
Could the authors report the method-specific hyperparameter search, training budget, and multi-seed results for the no-bypass baselines, and clarify how much of their performance degradation may arise from interface mismatch rather than bypass dependence?
We appreciate the reviewer's concern regarding baseline fairness. Our comparison preserves each method's original latent mechanism and training objectives while imposing a common no-bypass constraint, allowing us to evaluate visual competence retention under matched training conditions.
Specifically, each baseline generates latent representations through its original image-question forward pass. We preserve all latent tokens and their residual states across all 28 decoder layers, including method-specific boundary tokens and hybrid text–latent reasoning mechanisms. These representations construct the answer-side latent KV states without additional projection or pooling. As in CVRR, the original multimodal KV caches are removed, and question tokens use image-free KV caches from a text-only pass, ensuring that image-conditioned information reaches answer decoding through the preserved latent pathway.
The evaluated latent configurations and training objectives are summarized below.
| Method | Latent Representation | Latent Tokens () | Training Objectives |
|---|---|---|---|
| LVR | Last-position Hidden State | 8 | Answer CE + Visual Reconstruction MSE |
| Monet | Final-layer Hidden State | 10 | Answer CE + Layer-wise Cosine Alignment Loss |
| SkiLa | Sketch Hidden State | 27 | Answer CE + Latent Sketch Reconstruction Loss |
| Laser | Last-position Hidden State | Up to 8 | Answer CE + Dynamic Windowed Alignment Learning |
| UniVLR | Last-position Hidden State | 12 | Answer CE + Latent Alignment |
| HyLaR | Canvas Hidden State | Up to 32 | Answer CE + Canvas Prediction MSE |
Each baseline was initialized from its publicly released checkpoint rather than trained from scratch. All six baselines were further fine-tuned on the same Visual CoT dataset for three epochs using the same SFT configuration as CVRR: learning rate , AdamW, cosine scheduler, warmup ratio 0.05, effective batch size 128, and BF16 precision. The common SFT settings were standardized while preserving each method's original objectives, auxiliary losses, and loss weights. Evaluation used greedy decoding with each method's original generation configuration. We did not perform method-specific hyperparameter searches. Instead, we standardized the common SFT hyperparameters across all baselines and CVRR to ensure a fair and controlled comparison. Complete per-method settings will be provided in the revised appendix.
Under the same LoRA configuration (), each baseline uses 80.74M trainable LoRA parameters across 28 decoder blocks, compared with 2.88M in CVRR's single shared block. Thus, the baselines were not disadvantaged by a smaller trainable parameter budget.
We further examined the training loss, including method-specific auxiliary objectives where applicable, to assess optimization progress under the shared three-epoch schedule.
| Model | Epoch 1 | Epoch 2 | Epoch 3 |
|---|---|---|---|
| LVR | 3.0558 | 1.6915 | 1.6818 |
| Monet | 3.0295 | 1.2884 | 1.0246 |
| SkiLa | 8.3951 | 5.4534 | 2.6955 |
| Laser | 6.7041 | 2.6599 | 1.4224 |
| UniVLR | 6.8238 | 1.9757 | 0.8424 |
| HyLaR | 9.1048 | 6.6445 | 2.6363 |
| CVRR | 6.1702 | 1.7299 | 0.9307 |
All methods exhibit substantial reductions in training loss under the same training budget. For most methods, including CVRR, the loss reduction becomes smaller in the final epoch, suggesting that optimization is approaching stabilization.
Response to Reviewer UbSm (Part 3/3)
We additionally evaluated the no-bypass baselines across three independent training seeds (0, 45, 100) to examine the consistency of their performance.
| Model | Overall | MMVP Item | MMVP Pair | BLINK Overall | MME-RW-Lite |
|---|---|---|---|---|---|
| LVR-7B | 35.4 ± 1.3 | 49.9 ± 0.5 | 0.7 ± 0.7 | 37.8 ± 1.1 | 26.2 ± 1.1 |
| Monet-7B | 36.5 ± 1.6 | 50.6 ± 0.8 | 2.0 ± 0.7 | 38.5 ± 1.2 | 33.2 ± 1.2 |
| SkiLa-7B | 37.5 ± 1.3 | 29.7 ± 1.9 | 0.2 ± 0.4 | 38.5 ± 1.2 | 15.6 ± 1.4 |
| Laser-7B | 37.5 ± 1.1 | 50.0 ± 0.3 | 1.1 ± 0.8 | 38.1 ± 1.0 | 31.8 ± 0.9 |
| UniVLR-7B | 38.4 ± 0.8 | 49.9 ± 0.5 | 0.7 ± 0.7 | 38.9 ± 0.7 | 27.4 ± 1.3 |
| HyLaR-7B | 39.3 ± 1.9 | 42.2 ± 1.8 | 2.4 ± 1.0 | 33.5 ± 1.4 | 32.4 ± 1.5 |
| CVRR-7B | 81.5 ± 0.6 | 74.7 ± 0.7 | 52.7 ± 0.7 | 55.6 ± 0.4 | 46.9 ± 0.2 |
All values represent generation-based accuracy (%) as mean ± standard deviation across training seeds. These results are consistent with the single-run findings in Table 1, demonstrating that the limited visual competence retained by existing baselines under the strict interface is not specific to a single training seed. Moreover, most baselines obtain approximately 50% MMVP Item and near-zero MMVP Pair accuracy. Because the two images in each MMVP pair share the same question but have opposite answers, a decoder that receives no usable image information gives the same answer to both and therefore obtains exactly 50% Item and 0% Pair accuracy. Since the interface passes image-conditioned latent states to the decoder at every layer, this pattern suggests that the baselines may have difficulty effectively utilizing image-specific information from their latent representations under the strict interface. In contrast, CVRR achieves 81.5 ± 0.6% on Overall and 52.7 ± 0.7% on MMVP Pair Accuracy across the same three seeds, showing that strong visual competence can be retained under the same strict constraint.
Importantly, the weak latent-state reliance observed in Figure 1(a) was established under the original multimodal inference pathways, before any no-bypass modifications. Thus, this observation cannot be attributed to interface adaptation. Table 1 provides complementary evidence by evaluating visual competence retention after removing alternative image-conditioned pathways and retraining under the same constraint.
Together, these findings support our central observation that latent-state informativeness alone does not guarantee prediction reliance or visual competence retention. We will clarify the implementation details in the revised manuscript. PEARL was excluded because imposing the same constraint would require modifying its original predictive latent mechanism rather than merely removing an alternative multimodal pathway.
Gentle Reminder to Reviewer UbSm
Thank you again for your thoughtful review and valuable feedback. As the discussion period is approaching its end, we would like to kindly ask whether our responses and the additional experiments in the revised manuscript have addressed your concerns. If you have any remaining questions or concerns, we would be happy to provide further clarification.
Response to Reviewer UbSm (Part 1/3)
We sincerely appreciate the reviewer's constructive evaluation and recognition of our methodological contributions, particularly the distinction between latent-state informativeness and causal path necessity. We conducted additional experiments to address the concerns raised, as detailed below.
We appreciate the reviewer's insightful suggestion. To further distinguish sequential recurrent computation from independent feature refinement, we conducted both proposed controls: a Non-recurrent Bottleneck comparison and a Step-wise Visual Re-reading Intervention.
1. Non-recurrent Bottleneck Comparison
We introduced a Non-recurrent Bottleneck using the same native initialization h1, fixed visual evidence v, frozen decoder layer, and strict no-bypass interface as CVRR. While CVRR sequentially updates its recurrent state three times, the control processes the same h1 and v through three independent branches and aggregates their outputs. Each branch employs independently trained LoRA parameters with distinct scaling factors (α=8,16,32), allowing the branches to learn different visual transformations rather than sharing identical updates. The control was trained under the same recipe as CVRR. This enables a comparison of sequential state refinement against independent visual processing under comparable computational budgets, with the non-recurrent control using more trainable parameters.
The generation-based results are presented below.
Here, Δ denotes the accuracy improvement of CVRR in percentage points, with 95% CIs computed using paired bootstrap resampling (image-pair-clustered for MMVP).
CVRR outperforms the Non-recurrent Bottleneck across all metrics, achieving improvements of 4.2 percentage points on V∗ Overall and 6.7 percentage points on MMVP Pair Accuracy. The confidence intervals exclude zero for all metrics except Direct Attributes, providing evidence that sequential recurrent refinement offers additional benefits beyond independent visual processing.
2. Step-wise Visual Re-reading Ablation
We further examined the contribution of individual recurrent steps by selectively disabling question-to-visual attention at each recurrent update and independently retraining each ablation under the same T=4 training configuration. Step k refers to the k-th recurrent update, while native multimodal initialization h1 and the recurrent transition architecture are preserved across all conditions. All experiments used generation-based evaluation.
Here, Δ denotes the accuracy reduction relative to Full CVRR, with differences and paired bootstrap 95% CIs computed from unrounded sample-level results (pair-clustered resampling for MMVP). The All-off condition corresponds to the w/o Visual Re-read ablation in Table 2.
Despite independent retraining, disabling visual re-reading at any individual step reduces accuracy by 3.1–4.7 percentage points on V∗ and 6.7–10.0 percentage points on MMVP Pair Accuracy, with all confidence intervals excluding zero. Disabling visual access at all recurrent steps produces larger reductions of 12.1 and 34.7 percentage points, respectively.
Together with Figure 5, these results demonstrate the importance of visual access across multiple recurrent steps, while the Non-recurrent Bottleneck comparison supports the benefit of sequential state refinement. As the reviewer noted, path necessity is enforced by construction; our analyses focus on the computation performed along this required path. We will clarify our primary contribution as causal latent-state utilization and prediction-relevant iterative visual refinement.