Baixuan Li, Jialong Wu, Wenbiao Yin, Kuan Li, Zhongwang Zhang, Huifeng Yin, Zhengwei Tao, Liwen Zhang, Pengjun Xie, Jingren Zhou, Yong Jiang, Wentao Zhang, Zhiqiang Gao. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Transformers have demonstrated impressive capabilities across various tasks, yet their performance on compositional problems remains a subject of debate. In this study, we investigate the internal mechanisms underlying Transformers' behavior in compositional tasks. We find that complexity control strategies-particularly the choice of parameter initialization scale and weight decay-significantly influence whether the model learns primitive-level rules that generalize out-of-distribution (reasoning-based solutions) or relies solely on memorized mappings (memory-based solutions). By applying masking strategies to the model's information circuits and employing multiple complexity metrics, we reveal distinct internal working mechanisms associated with different solution types. Further analysis reveals that reasoning-based solutions exhibit a lower complexity bias, which aligns with the well-studied neuron condensation phenomenon. This lower complexity bias is hypothesized to be the key factor enabling these solutions to learn reasoning rules. We validate these conclusions across multiple real-world datasets, including image generation and natural language processing tasks, confirming the broad applicability of our findings.

Transformer-based Large Language Models (LLMs) have revolutionized Natural Language Processing by demonstrating exceptional performance across diverse tasks. This study investigates the impact of the parameter initialization scale on the training behavior and task preferences of LLMs. We discover that smaller initialization scales encourage models to favor reasoning tasks, whereas larger initialization scales lead to a preference for memorization tasks. We validate this reasoning bias via real datasets and meticulously designed anchor functions. Further analysis of initial training dynamics suggests that specific model components, particularly the embedding space and self-attention mechanisms, play pivotal roles in shaping these learning biases. We provide a theoretical framework from the perspective of model training dynamics to explain these phenomena. Additionally, experiments on real-world language tasks corroborate our theoretical insights. This work enhances our understanding of how initialization strategies influence LLM performance on reasoning tasks and offers valuable guidelines for training models.

Determining whether deep neural network (DNN) models can reliably recover target functions at overparameterization is a critical yet complex issue in the theory of deep learning. To advance understanding in this area, we introduce a concept we term “local linear recovery” (LLR), a weaker form of target function recovery that renders the problem more amenable to theoretical analysis. In the sense of LLR, we prove that functions expressible by narrower DNNs are guaranteed to be recoverable from fewer samples than model parameters. Specifically, we establish upper limits on the optimistic sample sizes, defined as the smallest sample size necessary to guarantee LLR, for functions in the space of a given DNN. Furthermore, we prove that these upper bounds are achieved in the case of two-layer tanh neural networks. Our research lays a solid groundwork for future investigations into the recovery capabilities of DNNs in overparameterized scenarios.

Dropout is a widely utilized regularization technique in the training of neural networks, nevertheless, its underlying mechanism and impact on achieving good generalization abilities remain to be further understood. In this work, we start by undertaking a rigorous theoretical derivation of the stochastic modified equations, with the primary aim of providing an effective approximation for the discrete iterative process of dropout. Meanwhile, we experimentally verify SDE's ability to approximate dropout under a wider range of settings. Subsequently, we empirically delve into the intricate mechanisms by which dropout facilitates the identification of flatter minima. This exploration is conducted through intuitive approximations, exploiting the structural analogies inherent in the Hessian of loss landscape and the covariance of dropout. Our empirical findings substantiate the ubiquitous presence of the Hessian-variance alignment relation throughout the training process of dropout.

Transformers have shown impressive capabilities across various tasks, but their performance on compositional problems remains a topic of debate. In this work, we investigate the mechanisms of how transformers behave on unseen compositional tasks. We discover that the parameter initialization scale plays a critical role in determining whether the model learns inferential (reasoning-based) solutions, which capture the underlying compositional primitives, or symmetric (memory-based) solutions, which simply memorize mappings without understanding the compositional structure. By analyzing the information flow and vector representations within the model, we reveal the distinct mechanisms underlying these solution types. We further find that inferential (reasoning-based) solutions exhibit low complexity bias, which we hypothesize is a key factor enabling them to learn individual mappings for single anchors. We validate our conclusions on various real-world datasets. Our findings provide valuable insights into the role of initialization scale in tuning the reasoning and memorizing ability and we propose the initialization rate $\gamma$ to be a convenient tunable hyper-parameter in common deep learning frameworks, where $1/d_{\mathrm{in}}^\gamma$ is the standard deviation of parameters of the layer with $d_{\mathrm{in}}$ input neurons.

It is important to understand how dropout, a popular regularization method, aids in achieving a good generalization solution during neural network training. In this work, we present a theoretical derivation of an implicit regularization of dropout, which is validated by a series of experiments. Additionally, we numerically study two implications of the implicit regularization, which intuitively rationalizes why dropout helps generalization. First, we find that input weights of hidden neurons tend to condense on isolated orientations trained with dropout. Condensation is a feature in the non-linear learning process, which makes the network less complex. Second, we find that the training with dropout leads to the neural network with a flatter minimum compared with standard gradient descent training, and the implicit regularization is the key to finding flat solutions. Although our theory mainly focuses on dropout used in the last hidden layer, our experiments apply to general dropout in training neural networks. This work points out a distinct characteristic of dropout compared with stochastic gradient descent and serves as an important basis for fully understanding dropout.

Side information has been demonstrated effective for recommendation by many existing works. However, even recent advancements can hardly extract the rich interactive knowledge and accurately model users' true interests due to ignorance of irregularities of reviews or temporal information, e.g., uneven time intervals. Therefore in this paper, we propose a Review-Enhanced Time-aware Sequential Recommendation framework (called RETSR) including review selector, long/short-term preferences extractors, etc. It can effectively exploit rich sequential, semantic, and temporal information to solve the aforementioned issues. More specifically, a review selector is first designed to select some important comments for obtaining users' long-term preferences and items' essential properties. Furthermore, users' short-term preferences are obtained by extracting meaningful time-sensitive features from two perspectives (i.e., reviews and interactions). Finally, extensive experiments show our framework significantly outperforms eleven baselines or state-of-the-art models on three popular real-world benchmarks, revealing its power for sequential recommendation tasks.

Understanding the structure of loss landscape of deep neural networks (DNNs) is obviously important. In this work, we prove an embedding principle that the loss landscape of a DNN "contains" all the critical points of all the narrower DNNs. More precisely, we propose a critical embedding such that any critical point, e.g., local or global minima, of a narrower DNN can be embedded to a critical point/affine subspace of the target DNN with higher degeneracy and preserving the DNN output function. Note that, given any training data, differentiable loss function and differentiable activation function, this embedding structure of critical points holds.This general structure of DNNs is starkly different from other nonconvex problems such as protein-folding.Empirically, we find that a wide DNN is often attracted by highly-degenerate critical points that are embedded from narrow DNNs. The embedding principle provides a new perspective to study the general easy optimization of wide DNNs and unravels a potential implicit low-complexity regularization during the training.Overall, our work provides a skeleton for the study of loss landscape of DNNs and its implication, by which a more exact and comprehensive understanding can be anticipated in the near future.