Assessing the impact the training data on machine learning models is crucial for understanding the behavior of the model, enhancing the transparency, and selecting training data. Influence function provides a theoretical framework for quantifying the effect of training data points on model's performance given a specific test data. However, the computational and memory costs of influence function presents significant challenges, especially for large-scale models, even when using approximation methods, since the gradients involved in computation are as large as the model itself. In this work, we introduce a novel approach that leverages dropout as a gradient compression mechanism to compute the influence function more efficiently. Our method significantly reduces computational and memory overhead, not only during the influence function computation but also in gradient compression process. Through theoretical analysis and empirical validation, we demonstrate that our method could preserves critical components of the data influence and enables its application to modern large-scale models.

Multi-label classification has broad applications and depends on powerful representations capable of capturing multi-label interactions. We introduce \textit{Diff-Feat}, a simple but powerful framework that extracts intermediate features from pre-trained diffusion-Transformer models for images and text, and fuses them for downstream tasks. We observe that for vision tasks, the most discriminative intermediate feature along the diffusion process occurs at the middle step and is located in the middle block in Transformer. In contrast, for language tasks, the best feature occurs at the noise-free step and is located in the deepest block. In particular, we observe a striking phenomenon across varying datasets: a mysterious "Layer $12$" consistently yields the best performance on various downstream classification tasks for images (under DiT-XL/2-256$\times$256). We devise a heuristic local-search algorithm that pinpoints the locally optimal "image-text"$\times$"block-timestep" pair among a few candidates, avoiding an exhaustive grid search. A simple fusion-linear projection followed by addition-of the selected representations yields state-of-the-art performance: 98.6\% mAP on MS-COCO-enhanced and 45.7\% mAP on Visual Genome 500, surpassing strong CNN, graph, and Transformer baselines by a wide margin. t-SNE and clustering metrics further reveal that \textit{Diff-Feat} forms tighter semantic clusters than unimodal counterparts. The code is available at https://github.com/lt-0123/Diff-Feat.

Multilingual Instruction Fine-Tuning (IFT) is essential for enabling large language models (LLMs) to generalize effectively across diverse linguistic and cultural contexts. However, the scarcity of high-quality multilingual training data and corresponding building method remains a critical bottleneck. While data selection has shown promise in English settings, existing methods often fail to generalize across languages due to reliance on simplistic heuristics or language-specific assumptions. In this work, we introduce Multilingual Data Quality and Diversity (M-DaQ), a novel method for improving LLMs multilinguality, by selecting high-quality and semantically diverse multilingual IFT samples. We further conduct the first systematic investigation of the Superficial Alignment Hypothesis (SAH) in multilingual setting. Empirical results across 18 languages demonstrate that models fine-tuned with M-DaQ method achieve significant performance gains over vanilla baselines over 60% win rate. Human evaluations further validate these gains, highlighting the increment of cultural points in the response. We release the M-DaQ code to support future research.

Existing decentralized stochastic optimization methods assume the lower-level loss function is strongly convex and the stochastic gradient noise has finite variance. These strong assumptions typically are not satisfied in real-world machine learning models. To address these limitations, we develop a novel decentralized stochastic bilevel optimization algorithm for the nonconvex bilevel optimization problem under heavy-tailed noises. Specifically, we develop a normalized stochastic variance-reduced bilevel gradient descent algorithm, which does not rely on any clipping operation. Moreover, we establish its convergence rate by innovatively bounding interdependent gradient sequences under heavy-tailed noises for nonconvex decentralized bilevel optimization problems. As far as we know, this is the first decentralized bilevel optimization algorithm with rigorous theoretical guarantees under heavy-tailed noises. The extensive experimental results confirm the effectiveness of our algorithm in handling heavy-tailed noises.

The manifold hypothesis suggests that high-dimensional data often lie on or near a low-dimensional manifold. Estimating the dimension of this manifold is essential for leveraging its structure, yet existing work on dimension estimation is fragmented and lacks systematic evaluation. This article provides a comprehensive survey for both researchers and practitioners. We review often-overlooked theoretical foundations and present eight representative estimators. Through controlled experiments, we analyze how individual factors such as noise, curvature, and sample size affect performance. We also compare the estimators on diverse synthetic and real-world datasets, introducing a principled approach to dataset-specific hyperparameter tuning. Our results offer practical guidance and suggest that, for a problem of this generality, simpler methods often perform better.

Speech is a rich signal, and labeled audio-text pairs are costly, making self-supervised learning essential for scalable representation learning. A core challenge in speech SSL is generating pseudo-labels that are both informative and efficient: strong labels, such as those used in HuBERT, improve downstream performance but rely on external encoders and multi-stage pipelines, while efficient methods like BEST-RQ achieve simplicity at the cost of weaker labels. We propose BiRQ, a bilevel SSL framework that combines the efficiency of BEST-RQ with the refinement benefits of HuBERT-style label enhancement. The key idea is to reuse part of the model itself as a pseudo-label generator: intermediate representations are discretized by a random-projection quantizer to produce enhanced labels, while anchoring labels derived directly from the raw input stabilize training and prevent collapse. Training is formulated as an efficient first-order bilevel optimization problem, solved end-to-end with differentiable Gumbel-softmax selection. This design eliminates the need for external label encoders, reduces memory cost, and enables iterative label refinement in an end-to-end fashion. BiRQ consistently improves over BEST-RQ while maintaining low complexity and computational efficiency. We validate our method on various datasets, including 960-hour LibriSpeech, 150-hour AMI meetings and 5,000-hour YODAS, demonstrating consistent gains over BEST-RQ.

Single-cell RNA-seq provides detailed molecular snapshots of individual cells but is notoriously noisy. Variability stems from biological differences, PCR amplification bias, limited sequencing depth, and low capture efficiency, making it challenging to adapt computational pipelines to heterogeneous datasets or evolving technologies. As a result, most studies still rely on principal component analysis (PCA) for dimensionality reduction, valued for its interpretability and robustness. Here, we improve upon PCA with a Random Matrix Theory (RMT)-based approach that guides the inference of sparse principal components using existing sparse PCA algorithms. We first introduce a novel biwhitening method, inspired by the Sinkhorn-Knopp algorithm, that simultaneously stabilizes variance across genes and cells. This enables the use of an RMT-based criterion to automatically select the sparsity level, rendering sparse PCA nearly parameter-free. Our mathematically grounded approach retains the interpretability of PCA while enabling robust, hands-off inference of sparse principal components. Across seven single-cell RNA-seq technologies and four sparse PCA algorithms, we show that this method systematically improves the reconstruction of the principal subspace and consistently outperforms PCA-, autoencoder-, and diffusion-based methods in cell-type classification tasks.

Standard Behavior Cloning (BC) fails to learn multimodal driving decisions, where multiple valid actions exist for the same scenario. W e explore Implicit Behavioral Cloning (IBC) with Energy-Based Models (EBMs) to better capture this multimodality. W e propose Data-Augmented IBC (DA-IBC), which improves learning by perturbing expert actions to form the counterexamples of IBC training and using better initialization for derivative-free inference. Experiments in the CARLA simulator with Bird's-Eye View inputs demonstrate that DA-IBC outperforms standard IBC in urban driving tasks designed to evaluate multimodal behavior learning in a test environment. The learned energy landscapes are able to represent multimodal action distributions, which BC fails to achieve.

Continual learning (CL) involves acquiring and accumulating knowledge from evolving tasks while alleviating catastrophic forgetting. Recently, leveraging contrastive loss to construct more transferable and less forgetful representations has been a promising direction in CL. Despite advancements, their performance is still limited due to confusion arising from both inter-task and intra-task features. To address the problem, we propose a simple yet effective contrastive strategy named \textbf{G}lobal \textbf{P}re-fixing, \textbf{L}ocal \textbf{A}djusting for \textbf{S}upervised \textbf{C}ontrastive learning (GPLASC). Specifically, to avoid task-level confusion, we divide the entire unit hypersphere of representations into non-overlapping regions, with the centers of the regions forming an inter-task pre-fixed \textbf{E}quiangular \textbf{T}ight \textbf{F}rame (ETF). Meanwhile, for individual tasks, our method helps regulate the feature structure and form intra-task adjustable ETFs within their respective allocated regions. As a result, our method \textit{simultaneously} ensures discriminative feature structures both between tasks and within tasks and can be seamlessly integrated into any existing contrastive continual learning framework. Extensive experiments validate its effectiveness.

Thermodynamic computing offers a potential route to energy-efficient computation. Unlike digital or quantum computing, which must at considerable energetic cost overpower or suppress thermal noise, thermodynamic computing is designed to use thermal noise as a source of energy. Physical devices whose states evolve under Langevin dynamics can be engineered to perform computations as they relax toward thermal equilibrium. Because these computations are carried out by the natural dynamics of the system, such devices can in principle operate with very low energy overhead, approaching fundamental thermodynamic limits [1-6]. A key challenge for thermodynamic computing is to identify algorithms that make efficient use of thermodynamic hardware and that reproduce the algebraic and machine-learning operations done digitally. Recent work has shown that thermodynamic computers can solve linear algebra problems, such as matrix inversion, in thermodynamic equilibrium [4, 5]. The advantage of equilibrium operation is that the computer's degrees of freedom obey the Boltzmann distribution, which depends in a precise way on the computer's potential energy. By choosing this potential energy appropriately, therefore, we can specify the desired computation.