弱监督

Synthetically-generated data plays an increasingly larger role in training large language models. However, while synthetic data has been found to be useful, studies have also shown that without proper curation it can cause LLM performance to plateau, or even "collapse", after many training iterations. In this paper, we formalize this question and develop a theoretical framework to investigate how much curation is needed in order to ensure that LLM performance continually improves. Our analysis is inspired by boosting, a classic machine learning technique that leverages a very weak learning algorithm to produce an arbitrarily good classifier. The approach we analyze subsumes many recently proposed methods for training LLMs on synthetic data, and thus our analysis sheds light on why they are successful, and also suggests opportunities for future improvement. We present experiments that validate our theory, and show that dynamically focusing labeling resources on the most challenging examples --- in much the same way that boosting focuses the efforts of the weak learner --- leads to improved performance.

Weak-to-strong generalization refers to the phenomenon where a stronger model trained under supervision from a weaker one can outperform its teacher. While prior studies aim to explain this effect, most theoretical insights are limited to abstract frameworks or linear/random feature models. In this paper, we provide a formal analysis of weak-to-strong generalization from a linear CNN (weak) to a two-layer ReLU CNN (strong). We consider structured data composed of label-dependent signals of varying difficulty and label-independent noise, and analyze gradient descent dynamics when the strong model is trained on data labeled by the pretrained weak model. Our analysis identifies two regimes—data-scarce and data-abundant—based on the signal-to-noise characteristics of the dataset, and reveals distinct mechanisms of weak-to-strong generalization. In the data-scarce regime, generalization occurs via benign overfitting or fails via harmful overfitting, depending on the amount of data, and we characterize the transition boundary. In the data-abundant regime, generalization emerges in the early phase through label correction, but we observe that overtraining can subsequently degrade performance.

In-context learning (ICL) enables large language models to perform few-shot learning by conditioning on labeled examples in the prompt. Despite its flexibility, ICL suffers from instability -- especially as prompt length increases with more demonstrations. To address this, we treat ICL as a source of weak supervision and propose a parameter-efficient method that disentangles demonstration-induced latent shifts from those of the query. An ICL-based teacher generates pseudo-labels on unlabeled queries, while a student predicts them using only the query input, updating a lightweight adapter. This captures demonstration effects in a compact, reusable form, enabling efficient inference while remaining composable with new demonstrations. Although trained on noisy teacher outputs, the student often outperforms its teacher through pseudo-label correction and coverage expansion, consistent with the weak-to-strong generalization effect. Empirically, our method improves generalization, stability, and efficiency across both in-domain and out-of-domain tasks, surpassing standard ICL and prior disentanglement methods.

Cross-modal representation learning aims to extract semantically aligned representations from heterogeneous modalities such as images and text. Existing multimodal VAE-based models often suffer from limited capability to align heterogeneous modalities or lack sufficient structural constraints to clearly separate the modality-specific and shared factors. In this work, we propose a novel framework, termed **D**isentangled **C**ross-**M**odal Representation Learning with **E**nhanced **M**utual Supervision (DCMEM). Specifically, our model disentangles the common and distinct information across modalities and regularizes the shared representation learned from each modality in a mutually supervised manner. Moreover, we incorporate the information bottleneck principle into our model to ensure that the shared and modality-specific factors encode exclusive yet complementary information. Notably, our model is designed to be trainable on both complete and partial multimodal datasets with a valid Evidence Lower Bound. Extensive experimental results demonstrate significant improvements of our model over existing methods on various tasks including cross-modal generation, clustering, and classification.

Weak-to-strong generalization—where a student model trained on imperfect labels generated by a weaker teacher nonetheless surpasses that teacher—has been widely observed, but the mechanisms that enable it have remained poorly understood. In this paper, through a theoretical analysis of simple models, we uncover three core mechanisms that can drive this phenomenon. First, by analyzing ridge linear regression, we study the interplay between the teacher and student regularization parameters and prove that a student can compensate for a teacher’s under-regularization and achieve lower test error. We also analyze the role of the parameterization regime of the models and show that qualitatively different phenomena can happen in different regimes. Second, by analyzing weighted ridge linear regression, we show that a student model with a regularization structure better aligned to the target function, can outperform its teacher. Third, in a nonlinear multi‐index learning setting, we demonstrate that a student can learn easy, task-specific features from the teacher while leveraging its own broader pre-training to learn hard‐to‐learn features that the teacher cannot capture.

Programmatic Weak Supervision (PWS) aims to reduce the cost of constructing large high quality labeled datasets often used in training modern machine learning models. A major component of the PWS pipeline is the label model, which amalgamates predictions from multiple noisy weak supervision sources, i.e. labeling functions (LFs), to label datapoints. While most label models are either probabilistic or adversarial, a recently proposed label model achieves strong empirical performance without falling into either camp. That label model constructs a polytope of plausible labelings using the LF predictions and outputs the "center" of that polytope as its proposed labeling. In this paper, we attempt to theoretically study that strategy by proposing Bayesian Balsubramani-Freund (BBF), a label model that implicitly constructs a polytope of plausible labelings and selects a labeling in its interior. We show an assortment of statistical results for BBF: log-concavity of its posterior, its form of solution, consistency, and rates of convergence. Extensive experiments compare our proposed method against twelve baseline label models over eleven datasets. BBF compares favorably to other Bayesian label models and label models that don't use datapoint features -- matching or exceeding their performance on eight out of eleven datasets.

As future superhuman models become increasingly complex, accurately supervising their behavior may exceed human capabilities. Recent works have demonstrated that in such scenarios, weak models can effectively supervise strong models, a phenomenon known as weak-to-strong generalization. However, we find that naive weak-to-strong generalization fails under distribution shifts, often leading to worse performance of the strong model than its weak supervisors. To address this, we propose RAVEN, a robust weak-to-strong generalization framework that dynamically learns the optimal combinations of weak models in addition to parameters of the strong model. We demonstrate the effectiveness of RAVEN on image classification, text classification, and preference alignment tasks. RAVEN outperforms alternative baselines by over 30% on out-of-distribution tasks while matching or surpassing existing methods on in-distribution tasks. Moreover, our results show that RAVEN assigns higher weights to more accurate weak models, demonstrating its ability to automatically identify trustworthy supervision.

Verifiers can improve language model (LM) capabilities by providing feedback or selecting the best response from a pool of generated candidates. Currently, high-quality verifiers are either unscalable (e.g., humans) or limited in utility (e.g., tools like Lean for formal proofs). While LM judges and reward models have become broadly useful as general-purpose verifiers, a significant performance gap remains between them and oracle verifiers. To help close this gap, we introduce Weaver, a framework for designing a strong verifier by combining multiple weak, imperfect verifiers. First we find that weighted ensembles of verifiers, which typically require learning from labeled data, significantly outperform unweighted combinations due to differences in the verifiers. To reduce the dependency on labeled data, Weaver leverages weak supervision to estimate each verifier’s accuracy and combines their outputs into a unified score that better reflects true response quality. However, directly applying weak supervision algorithms poses several challenges, including inconsistent verifier output formats and handling low-quality verifiers. Weaver addresses these challenges by using dataset statistics to normalize outputs and filter specific verifiers. We study the effectiveness of Weaver in repeated sampling settings, where a model generates multiple candidate responses at test time and a verifier is used to select the correct one. Our evaluations demonstrate that Weaver significantly improves the pass@1 performance across several reasoning and math tasks, achieving o3-mini level accuracy with Llama 3.3 70B Instruct (a much cheaper non-reasoning model) as the generator, and an ensemble of smaller judge and reward models as the verifiers (86.2% average). This gain mirrors the jump achieved between GPT-4o and o3-mini (69.0% vs. 86.7%), which required extensive finetuning and post-training interventions. To make Weaver more efficient, we train a compact 400M cross-encoder using Weaver's combined output scores. This distilled model retains 98.7% of Weaver's full accuracy while reducing verification compute by up to 99.97%.

Boosting is a key method in statistical learning, allowing for converting weak learners into strong ones. While well studied in the realizable case, the statistical properties of weak-to-strong learning remain less understood in the agnostic setting, where there are no assumptions on the distribution of the labels. In this work, we propose a new agnostic boosting algorithm with substantially improved sample complexity compared to prior works under very general assumptions. Our approach is based on a reduction to the realizable case, followed by a margin-based filtering of high-quality hypotheses. Furthermore, we show a nearly-matching lower bound, settling the sample complexity of agnostic boosting up to logarithmic factors.

We study the problem of regression with interval targets, where only upper and lower bounds on target values are available in the form of intervals. This problem arises when the exact target label is expensive or impossible to obtain, due to inherent uncertainties. In the absence of exact targets, traditional regression loss functions cannot be used. First, we study the methodology of using a loss functions compatible with interval targets, for which we establish non-asymptotic generalization bounds based on smoothness of the hypothesis class that significantly relaxing prior assumptions of realizability and small ambiguity degree. Second, we propose a novel min-max learning formulation: minimize against the worst-case (maximized) target labels within the provided intervals. The maximization problem in the latter is non-convex, but we show that good performance can be achieved with the incorporation of smoothness constraints. Finally, we perform extensive experiments on real-world datasets and show that our methods achieve state-of-the-art performance.

This paper studies the problem of learning hash codes from noisy supervision, which is a practical yet challenging task. This problem is important in extensive real-world applications such as image retrieval and cross-modal retrieval. However, most of the existing methods focus on label denoising to address this problem, but ignore the geometric structure of the hash space, which is critical for learning stable hash codes. Towards this end, this paper proposes a novel framework named Semantic Geometry Shaping (SEGA) that explicitly refines the semantic geometry of hash space. Specifically, we first learn dynamic class prototypes as semantic anchors and cluster hash embeddings around these prototypes to keep structural stability. We then leverage both the energy of predicted distributions and structure-based divergence to estimate the uncertainty of instances and calibrate the supervision in a soft manner. Moreover, we introduce structure-aware interpolation to improve the class boundaries. To verify the effectiveness of our design, we give the theoretical analysis for the proposed framework. Experiments on a range of widely-used retrieval datasets justify the superiority of our SEGA over extensive strong baselines under noisy supervision.

Visual Place Recognition (VPR) enables robust localization through image retrieval based on learned descriptors. However, drastic appearance variations of images at the same place caused by viewpoint changes can lead to inconsistent supervision signals, thereby degrading descriptor learning. Existing methods either rely on manually defined cropping rules or labeled data for view differentiation, but they suffer from two major limitations: (1) reliance on labels or handcrafted rules restricts generalization capability; (2) even within the same view direction, occlusions can introduce feature ambiguity. To address these issues, we propose MutualVPR, a mutual learning framework that integrates unsupervised view self-classification and descriptor learning. We first group images by geographic coordinates, then iteratively refine the clusters using K-means to dynamically assign place categories without manual labeling. Specifically, we adopt a DINOv2-based encoder to initialize the clustering. During training, the encoder and clustering co-evolve, progressively separating drastic appearance variations of the same place and enabling consistent supervision. Furthermore, we find that capturing fine-grained image differences at a place enhances robustness. Experiments demonstrate that MutualVPR achieves state-of-the-art (SOTA) performance across multiple datasets, validating the effectiveness of our framework in improving view direction generalization, occlusion robustness.

Existing machine-generated text (MGT) detection methods implicitly assume labels as the "golden standard". However, we reveal boundary ambiguity in MGT detection, implying that traditional training paradigms are inexact. Moreover, limitations of human cognition and the superintelligence of detectors make inexact learning widespread and inevitable. To this end, we propose an easy-to-hard enhancement framework to provide reliable supervision under such inexact conditions. Distinct from knowledge distillation, our framework employs an easy supervisor targeting relatively simple longer-text detection tasks (despite weaker capabilities), to enhance the more challenging target detector. Firstly, longer texts targeted by supervisors theoretically alleviate the impact of inexact labels, laying the foundation for reliable supervision. Secondly, by structurally incorporating the detector into the supervisor, we theoretically model the supervisor as a lower performance bound for the detector. Thus, optimizing the supervisor indirectly optimizes the detector, ultimately approximating the underlying "golden" labels. Extensive experiments across diverse practical scenarios, including cross-LLM, cross-domain, mixed text, and paraphrase attacks, demonstrate the framework's significant detection effectiveness. The code is available at: \url{https://github.com/tmlr-group/Easy2Hard}.

Learning from Label Proportions (LLP) is a weakly-supervised paradigm that uses bag-level label proportions to train instance-level classifiers, offering a practical alternative to costly instance-level annotation. However, the weak supervision makes effective training challenging, and existing methods often rely on pseudo-labeling, which introduces noise. To address this, we propose RLPL, a two-stage framework. In the first stage, we use unsupervised contrastive learning to pretrain the encoder and train an auxiliary classifier with bag-level supervision. In the second stage, we introduce an LLP-OTD mechanism to refine pseudo labels and split them into high- and low-confidence sets. These sets are then used in LLPMix to train the final classifier. Extensive experiments and ablation studies on multiple benchmarks demonstrate that RLPL achieves comparable state-of-the-art performance and effectively mitigates pseudo-label noise.

Pre-trained encoders for offline feature extraction followed by multiple instance learning (MIL) aggregators have become the dominant paradigm in computational pathology (CPath), benefiting cancer diagnosis and prognosis. However, performance limitations arise from the absence of encoder fine-tuning for downstream tasks and disjoint optimization with MIL. While slide-level supervised end-to-end (E2E) learning is an intuitive solution to this issue, it faces challenges such as high computational demands and suboptimal results. These limitations motivate us to revisit E2E learning. We argue that prior work neglects inherent E2E optimization challenges, leading to performance disparities compared to traditional two-stage methods. In this paper, we pioneer the elucidation of optimization challenge caused by sparse-attention MIL and propose a novel MIL called ABMILX. ABMILX mitigates this problem through global correlation-based attention refinement and multi-head mechanisms. With the efficient multi-scale random patch sampling strategy, an E2E trained ResNet with ABMILX surpasses SOTA foundation models under the two-stage paradigm across multiple challenging benchmarks, while remaining computationally efficient ($<$ 10 RTX3090 GPU hours). We demonstrate the potential of E2E learning in CPath and calls for greater research focus in this area. The code is https://github.com/DearCaat/E2E-WSI-ABMILX.