In particular, prior works often generate pseudo-label for unlabeled images and then train the model on those samples.

Gradient descent and its variants are de facto standard algorithms for training machine learning models. As gradient descent is sensitive to its hyperparame-ters, we need to tune the hyperparameters carefully using a grid search.

Privacy is a growing concern in modern deep-learning systems and applications. Differentially private (DP) training prevents the leakage of sensitive information in the collected training data from the trained machine learning models. DP op-timizers, including DP stochastic gradient descent (DPSGD) and its variants, privatize the training procedure by gradient clipping and DP noise injection. However, in practice, DP models trained using DPSGD and its variants often suffer from significant model performance degradation. Such degradation prevents the application of DP optimization in many key tasks, such as foundation model pre-training.

In online advertising, heterogeneous advertiser requirements give rise to numerous customized bidding tasks that are typically optimized independently, resulting in extensive computation and limited data efficiency. Multi-task learning offers a principled framework to train these tasks jointly through shared representations. However, existing multi-task optimization strategies are primarily guided by training dynamics and often generalize poorly in volatile bidding environments. To this end, we present Validation-Aligned Multi-task Optimization (VAMO), which adaptively assigns task weights based on the alignment between per-task training gradients and a held-out validation gradient, thereby steering updates toward validation improvement and better matching deployment objectives. We further equip the framework with a periodicity-aware temporal module and couple it with an advanced generative auto-bidding backbone to enhance cross-task transfer of seasonal structure and strengthen bidding performance. Meanwhile, we provide theoretical insights into the proposed method, e.g., convergence guarantee and alignment analysis. Extensive experiments on both simulated and large-scale real-world advertising systems consistently demonstrate significant improvements over typical baselines, illuminating the effectiveness of the proposed approach.

Dropout is a standard training technique for neural networks that consists of randomly deactivating units at each step of their gradient-based training. It is known to improve performance in many settings, including in the large-scale training of language or vision models. As a first step towards understanding the role of dropout in large neural networks, we study the large-width asymptotics of gradient descent with dropout on two-layer neural networks with the mean-field initialization scale. We obtain a rich asymptotic phase diagram that exhibits five distinct nondegenerate phases depending on the relative magnitudes of the dropout rate, the learning rate, and the width. Notably, we find that the well-studied "penalty" effect of dropout only persists in the limit with impractically small learning rates of order $O(1/\text{width})$. For larger learning rates, this effect disappears and in the limit, dropout is equivalent to a "random geometry" technique, where the gradients are thinned randomly after the forward and backward pass have been computed. In this asymptotic regime, the limit is described by a mean-field jump process where the neurons' update times follow independent Poisson or Bernoulli clocks (depending on whether the learning rate vanishes or not). For some of the phases, we obtain a description of the limit dynamics both in path-space and in distribution-space. The convergence proofs involve a mix of tools from mean-field particle systems and stochastic processes. Together, our results lay the groundwork for a renewed theoretical understanding of dropout in large-scale neural networks.

Reinforcement Learning with Human Feedback (RLHF) has been the dominant approach for improving the reasoning capabilities of Large Language Models (LLMs). Recently, Reinforcement Learning with Verifiable Rewards (RLVR) has simplified this paradigm by replacing the reward and value models with rule-based verifiers. A prominent example is Group Relative Policy Optimization (GRPO). However, GRPO inherently suffers from a length bias, since the same advantage is uniformly assigned to all tokens of a response. As a result, longer responses distribute the reward over more tokens and thus contribute disproportionately to gradient updates. Several variants, such as DAPO and Dr. GRPO, modify the token-level aggregation of the loss, yet these methods remain heuristic and offer limited interpretability regarding their implicit token preferences. In this work, we explore the possibility of allowing the model to learn its own token preference during optimization. We unify existing frameworks under a single formulation and introduce a learnable parameter $λ$ that adaptively controls token-level weighting. We use $λ$-GRPO to denote our method, and we find that $λ$-GRPO achieves consistent improvements over vanilla GRPO and DAPO on multiple mathematical reasoning benchmarks. On Qwen2.5 models with 1.5B, 3B, and 7B parameters, $λ$-GRPO improves average accuracy by $+1.9\%$, $+1.0\%$, and $+1.7\%$ compared to GRPO, respectively. Importantly, these gains come without any modifications to the training data or additional computational cost, highlighting the effectiveness and practicality of learning token preferences.

Differential privacy (DP) is a prominent method for protecting information about individuals during data analysis. Training neural networks with differentially private stochastic gradient descent (DPSGD) influences the model's learning dynamics and, consequently, its output. This can affect the model's performance and fairness. While the majority of studies on the topic report a negative impact on fairness, it has recently been suggested that fairness levels comparable to non-private models can be achieved by optimizing hyperparameters for performance directly on differentially private models (rather than re-using hyperparameters from non-private models, as is common practice). In this work, we analyze the generalizabil-ity of this claim by 1) comparing the disparate impact of DPSGD on different performance metrics, and 2) analyzing it over a wide range of hyperparameter settings. We highlight that a disparate impact on one metric does not necessarily imply a disparate impact on another. Most importantly, we show that while optimizing hyperparameters directly on differentially private models does not mitigate the disparate impact of DPSGD reliably, it can still lead to improved utility-fairness trade-offs compared to re-using hyperparameters from non-private models. We stress, however, that any form of hyperparameter tuning entails additional privacy leakage, calling for careful considerations of how to balance privacy, utility and fairness. Finally, we extend our analyses to DPSGD-Global-Adapt, a variant of DPSGD designed to mitigate the disparate impact on accuracy, and conclude that this alternative may not be a robust solution with respect to hyperparameter choice.

Adam has been shown to outperform gradient descent on large language models by a larger margin than on other tasks, but it is unclear why. We show that a key factor in this performance gap is the heavy-tailed class imbalance found in language tasks.

In particular, its convergence guarantees can be obtained under relative smoothness and convexity of the Fenchel transform of the potential, with respect to the objective.