- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper theoretically characterizes the emergence of an implicit curriculum in a stylized shared-attention Transformer, showing that token selection is learned before XOR composition despite simultaneous training on both tasks.
The authors claimed that in a simplified model with shared attention and task-specific ReLU heads, the token selection would reach lower test errors before the XOR composition.
Then, they also gives a lower bound for XOR-only training under same training budgets and explains the ordering.
Finally, the three experiments (Synthetic data, MNIST sequences, LLM checkpoints) support the main prediction, but the evidence becomes less mechanistically conclusive as the setting becomes more realistic.
Critical strengths
-
The theoretical analysis provides a learning-order guarantee under simultaneous training and the theoretical derivations are clearly presented and carefully structured.
-
The experiments complement the theoretical analysis by demonstrating the predicted learning order on synthetic and MNIST-based tasks and the results partially support the assumption.
Overall, this is an interesting paper. Although the theoretical analysis relies on relatively strong assumptions, the experiments show that the proposed method is effective to some extent. If the authors can further improve the completeness of the paper and extend the method to widely used llm, the paper could become a promising line of work worth following.
Critical weaknesses
- The key weakness is the gap of theory and experiment(especially contemporary large language models).
The theory establishes a concrete mechanism for implicit curriculum, but relies on restrictive structural and scaling assumptions, particularly an explicitly shared relevance signal and a reduced multi-task architecture. Consequently, its applicability to modern LLM pretraining remains uncertain. Pythia-6.9B, Amber-7B, OLMo-2-7B the experiments used is totally outside the theoretical setting and the author also do not provide theoretical anaslyation for the LLM checkpoints.
- The experimental section is brief with insufficient analysis without ablation study.
The experiments mainly show the predicted learning order, but they do not clearly test the reason behind it. In particular, the LLM experiments are quite different from the theoretical setting. Similar learning curves alone cannot prove that the proposed mechanism works in real LLMs. More controlled experiments and ablation studies would help connect the theory to the experimental results.
For example, have the authors considered controlled comparisons between ReLU and SwiGLU to test the robustness? More broadly, how sensitive is this behavior to architectural choices.
I understand that space is limited. The authors could reorganize the paper, for example, by moving some less important equations to the appendix.
Questions for authors
As noted in the Weaknesses section.
The paper appears to be primarily theoretical, as the experimental section is limited in both scope and depth of analysis, however, two of the four contributions listed at the end of the intro are about experiments and the authors' main insight comes from the gap between the experimental observations and the theory, therefore, I believe the authors need to explain this key weakness more clearly.
5 comments
Response to Reviewer DX5S (Part 2/2)
W2.2: Have the authors considered controlled comparisons between ReLU and SwiGLU to test the robustness? More broadly, how sensitive is this behavior to architectural choices.
Thanks for the suggestion on architectural choices experiments! We expanded the comparison to 12 configurations across three tasks, with five seeds per setting. Across all 180 runs, perception reached sustained 90% balanced accuracy before parity. This ordering persists under changes to activation functions, attention, depth, normalization, pooling, and the training objective. These results suggest that the perception-before-parity learning order we study extends to a broader range of architectures and training objectives.
| Variant | Synthetic XOR | MNIST: 2 signals | MNIST: 3 signals |
|---|---|---|---|
| ReLU baseline | 8.8 ± 2.7 | 90.0 ± 23.5 | 462.0 ± 108.0 |
| SwiGLU FFN | 11.2 ± 2.3 | 70.0 ± 21.2 | 399.0 ± 124.9 |
| GELU FFN + GELU head | 17.6 ± 7.9 | 106.0 ± 33.6 | 410.0 ± 80.1 |
| 1 layer | 20.0 ± 2.4 | 120.0 ± 31.6 | 873.0 ± 153.1 |
| 4 layers | 9.6 ± 3.8 | 92.0 ± 25.9 | 429.0 ± 76.2 |
| Linear-attention blocks | 8.4 ± 3.3 | 176.0 ± 25.1 | 1568.0 ± 461.5 |
| Mean pooling | 8.8 ± 2.7 | 90.0 ± 23.5 | 474.0 ± 153.6 |
| Squared loss | 8.0 ± 2.8 | 84.0 ± 20.7 | 437.0 ± 110.8 |
| Post-LayerNorm | 25.2 ± 15.8 | 40.0 ± 7.1 | 134.0 ± 8.9 |
| RMSNorm | 10.4 ± 1.7 | 90.0 ± 15.8 | 402.0 ± 53.2 |
| Linear attention + mean pooling | 4.0 ± 1.4 | 184.0 ± 27.0 | 1568.0 ± 333.3 |
| SwiGLU + SiLU head + squared loss | 14.8 ± 3.0 | 70.0 ± 12.2 | 419.0 ± 73.1 |
We also appreciate the reviewer’s suggestion to move some less important derivations and technical details to appendix and save some space for the experiment results. We add these results in Table 1 in the revised manuscript.
References
[1] Chen et al. Distributional Associations vs In-Context Reasoning: A Study of Feed-forward and Attention Layers. ICLR, 2025.
[2] Panigrahi et al. Progressive distillation induces an implicit curriculum. ICLR, 2025.
[3] Shi et al. Why Larger Language Models Do In-context Learning Differently? ICML, 2024.
[4] Abbe et al. Provable advantage of curriculum learning on parity targets with mixed inputs. NeurIPS, 2023.
[5] Wang et al. Learning Compositional Functions with Transformers from Easy-to-Hard Data. COLT, 2025.
[6] Kim et al. Transformers Provably Solve Parity Efficiently with Chain of Thought. ICLR, 2025.
Thank you for the responses and additional experiments. The new ablation study strengthen the evidence and the architectureal comparision support the robustness of the observation of learning order.
However, my main concern remains the connection between the machanism and the mordern LLM experiments. The additional results do not establish that the same machanism explain explains the observaion in modern LLMs
I appreciate the clarification of the theory’s scope, but the role of the LLM experiments and the scope of the empirical contributions still need clearer justification. Therefore, I will maintain my original score.
Response to Reviewer DX5S (Part 1/2)
We appreciate the reviewer’s thoughtful comments and provide our responses below.
W1: The key weakness is the gap of theory and experiment(especially contemporary large language models).
Thanks for raising this! We acknowledge that our theorem does not directly characterize the training dynamics of modern LLMs, and we do not claim it does. Our goal is to keep the key structure that matters for an implicit curriculum and drop the rest, so we proposed the stylized problem to capture the training dynamic of a shared-attention multi-task Transformer. What we keep: both tasks are trained together from the first step under a fixed mixture, attention and the ReLU heads are updated at every iteration, and the two tasks share the need to locate the relevant tokens (the shared relevance signal). Prior theory on this question freezes one part, trains in stages, or imposes an explicit curriculum [4,5,6], so on this point our setting is closer to the actual empirical setting.
A simple model also lets us pin down the cause. Theorem 4.2 shows that without the selection data, XOR stays near chance under the same budget. Such a statement cannot be proved, or even tested, on a 7B model, since the end-to-end training dynamics of deep Transformers are still open problems. This is also common practice in closely related theory work [1,2,3], which pairs simplified theory with experiments on Pythia, BERT and Llama.
W2.1: The experimental section is brief with insufficient analysis without ablation study. Similar learning curves alone cannot prove that the proposed mechanism works in real LLMs.
Thank you. We would first like to clarify that our experiments are intended as an extension of our theoretical analysis, exploring whether the proposed mechanism and phenomenon may generalize to a broader range of scenarios beyond those covered by our theory. For the experiment conducted under the theoretical settings, we refer the reviewer to the section Parameters That Match Theoretical Settings in our global rebuttal.
To test the reason behind the ordering, and not only the ordering itself, we added two controlled experiments (Global Rebuttal).
(1) Results for weak selection. With data, initialization, learning rate, attention cap, and update budget all matched, joint training reaches 95.08 ± 1.32% XOR accuracy, composition-only training reaches 60.40 ± 1.01%, and a frozen uniform query reaches 57.74 ± 1.51%. Removing the selection data removes most of the XOR learning.
(2) Ablation on loss component ratio. On binary MNIST, with the composition objective and data fixed, composition accuracy stays at chance for selection-loss weight β≤0.01 and reaches about 90% for β≥0.05 within the same 6,000 updates.
In addition, freezing the attention on MNIST keeps composition at chance (51.39 ± 0.79%, Figure 5).