Fan Wei

Fan Wei, Runmin Dong, Yushan Lai et al.

Diffusion-based remote sensing (RS) generative foundation models are cruial for downstream tasks. However, these models rely on large amounts of globally representative data, which often contain redundancy, noise, and class imbalance, reducing training efficiency and preventing convergence. Existing RS diffusion foundation models typically aggregate multiple classification datasets or apply simplistic deduplication, overlooking the distributional requirements of generation modeling and the heterogeneity of RS imagery. To address these limitations, we propose a training-free, two-stage data pruning approach that quickly select a high-quality subset under high pruning ratios, enabling a preliminary foundation model to converge rapidly and serve as a versatile backbone for generation, downstream fine-tuning, and other applications. Our method jointly considers local information content with global scene-level diversity and representativeness. First, an entropy-based criterion efficiently removes low-information samples. Next, leveraging RS scene classification datasets as reference benchmarks, we perform scene-aware clustering with stratified sampling to improve clustering effectiveness while reducing computational costs on large-scale unlabeled data. Finally, by balancing cluster-level uniformity and sample representativeness, the method enables fine-grained selection under high pruning ratios while preserving overall diversity and representativeness. Experiments show that, even after pruning 85\% of the training data, our method significantly improves convergence and generation quality. Furthermore, diffusion foundation models trained with our method consistently achieve state-of-the-art performance across downstream tasks, including super-resolution and semantic image synthesis. This data pruning paradigm offers practical guidance for developing RS generative foundation models.

Bin Luo, Runmin Dong, Zhaoyang Luo et al.

Diffusion-based methods demonstrate significant potential for remote sensing image super-resolution at large scaling factors, particularly in reference-based super-resolution (RefSR) where high-resolution reference images provide critical fine-grained texture priors. However, existing methods often suffer from a trade-off between over-reliance on reference information, which leads to texture artifacts, and underutilization, which results in insufficient detail recovery. To address these issues, we propose DS-DiT, a Decoupled Siamese Diffusion Transformer method that decouples low-resolution and reference interactions at the attention level. By enabling low-resolution structural priors and reference texture information to interact independently with the noisy latent, the framework effectively mitigates inter-source competition. Furthermore, to compensate for the limited local modeling ability of global attention, we introduce a Patch-Level Weights (PLW) module that adaptively modulates the fusion of conditional sources. In addition, this siamese architecture facilitates an autoguidance strategy during inference, which enhances reconstruction by exploiting the prediction discrepancy between strong and weak reference conditions. This approach boosts generation quality without additional training. Experimental results across multiple datasets and scaling factors demonstrate that DS-DiT outperforms existing methods in both quantitative metrics and visual fidelity.

Jon Kleinberg, Fan Wei

The formalism of language generation in the limit studies generative models by requiring an algorithm, given strings from a hidden true language, to eventually generate new valid strings. A core issue is the tension between validity and breadth. Prior work quantified breadth via asymptotic density, where the priority is generating strings early in a natural countable ordering. Here, we study density when the strings are embedded in $d$ dimensions, a ubiquitous structure in current generative models. Our goal is for the generated strings to be dense throughout the embedding. This requires a different measure, the Banach density, which captures whether a set contains large sparse regions. Using Banach density uncovers a rich structure based on dimension and the topology of the language collection. We prove that in dimension one, when the underlying topological space has finite Cantor-Bendixson rank, an algorithm can always generate a subset of the true language with an optimal lower Banach density of 1/2. However, for collections with infinite Cantor-Bendixson rank, there are cases where no algorithm can achieve any positive lower Banach density; the generated set must contain arbitrarily large, sparse regions. This reveals a topological contrast unseen with asymptotic density, where 1/2 is always achievable. We also extend our results to a family of measures interpolating between Banach and asymptotic density. Finally, in dimension $d \geq 2$, our positive result for Banach density encounters a Ramsey-theoretic obstacle regarding two-colored point sets. Overcoming this requires a nondegeneracy condition: the embedding of the true language must be sufficiently represented throughout the full $d$-dimensional space.

Jinxiao Zhang, Runmin Dong, Xiyong Wu et al.

Earth observation is becoming one of the largest data-producing activities in science, yet current pipelines still treat compression as a storage and transmission tool rather than a new way to use data. We present a generative compression framework that learns from historical Earth observation archives and enables on-demand 100x to 10,000x data reduction across downstream tasks. Unlike general visual data, Earth observation repeatedly measures the same evolving planet, making historical-prior learning feasible for extreme compression. To realize this paradigm, we train large generative compression models at exascale on the LineShine Armv9 CPU supercomputer, with co-optimization across model design, kernels, memory hierarchy, runtime, and parallelism. Our implementation sustains 1.54 EFLOP/s and peaks at 2.16 EFLOP/s in end-to-end training. This work shows that historical-prior generative compression can turn Earth observation data into an active, task-adaptive foundation for acquisition, delivery, storage, and scientific use.

Jon Kleinberg, Fan Wei

The success of large language models (LLMs) has motivated formal theories of language generation and learning. We study the framework of \emph{language generation in the limit}, where an adversary enumerates strings from an unknown language $K$ drawn from a countable class, and an algorithm must generate unseen strings from $K$. Prior work showed that generation is always possible, and that some algorithms achieve positive lower density, revealing a \emph{validity--breadth} trade-off between correctness and coverage. We resolve a main open question in this line, proving a tight bound of $1/2$ on the best achievable lower density. We then strengthen the model to allow \emph{partial enumeration}, where the adversary reveals only an infinite subset $C \subseteq K$. We show that generation in the limit remains achievable, and if $C$ has lower density $α$ in $K$, the algorithm's output achieves density at least $α/2$, matching the upper bound. This generalizes the $1/2$ bound to the partial-information setting, where the generator must recover within a factor $1/2$ of the revealed subset's density. We further revisit the classical Gold--Angluin model of \emph{language identification} under partial enumeration. We characterize when identification in the limit is possible -- when hypotheses $M_t$ eventually satisfy $C \subseteq M \subseteq K$ -- and in the process give a new topological formulation of Angluin's characterization, showing that her condition is precisely equivalent to an appropriate topological space having the $T_D$ separation property.

Jon Kleinberg, Fan Wei

The recent successes of large language models (LLMs) have led to a surge of theoretical research into language generation. A recent line of work proposes an abstract view, called language generation in the limit, where generation is seen as a game between an adversary and an algorithm: the adversary generates strings from an unknown language $K$, chosen from a countable collection of candidate languages, and after seeing a finite set of these strings, the algorithm must generate new strings from $K$ that it has not seen before. This formalism highlights a key tension: the trade-off between validity (the algorithm should only produce strings from the language) and breadth (it should be able to produce many strings from the language). This trade-off is central in applied language generation as well, where it appears as a balance between hallucination (generating invalid utterances) and mode collapse (generating only a restricted set of outputs). Despite its importance, this trade-off has been challenging to study quantitatively. We develop ways to quantify this trade-off by formalizing breadth using measures of density. Existing algorithms for language generation in the limit produce output sets that can have zero density in the true language, and this important failure of breadth might seem unavoidable. We show, however, that such a failure is not necessary: we provide an algorithm for language generation in the limit whose outputs have strictly positive density in $K$. We also study the internal representations built by these algorithms, specifically the sequence of hypothesized candidate languages they consider, and show that achieving the strongest form of breadth may require oscillating indefinitely between high- and low-density representations. Our analysis introduces a novel topology on language families, with notions of convergence and limit points playing a key role.

Zhenkun Shi, Jun Zhu, Dehang Wang et al.

A key challenge in enzyme annotation is identifying the biochemical reactions catalyzed by proteins. Most existing methods rely on Enzyme Commission (EC) numbers as intermediaries: they first predict an EC number and then retrieve the associated reactions. This indirect strategy introduces ambiguity due to the complex many-to-many mappings among proteins, EC numbers, and reactions, and is further complicated by frequent updates to EC numbers and inconsistencies across databases. To address these challenges, we present RXNRECer, a transformer-based ensemble framework that directly predicts enzyme-catalyzed reactions without relying on EC numbers. It integrates protein language modeling and active learning to capture both high-level sequence semantics and fine-grained transformation patterns. Evaluations on curated cross-validation and temporal test sets demonstrate consistent improvements over six EC-based baselines, with gains of 16.54% in F1 score and 15.43% in accuracy. Beyond accuracy gains, the framework offers clear advantages for downstream applications, including scalable proteome-wide reaction annotation, enhanced specificity in refining generic reaction schemas, systematic annotation of previously uncurated proteins, and reliable identification of enzyme promiscuity. By incorporating large language models, it also provides interpretable rationales for predictions. These capabilities make RXNRECer a robust and versatile solution for EC-free, fine-grained enzyme function prediction, with potential applications across multiple areas of enzyme research and industrial applications.

Yi Zhang, Fan Wei, Jingyi Li et al.

The use of children's drawings to examining their conceptual understanding has been proven to be an effective method, but there are two major problems with previous research: 1. The content of the drawings heavily relies on the task, and the ecological validity of the conclusions is low; 2. The interpretation of drawings relies too much on the subjective feelings of the researchers. To address this issue, this study uses the Large Language Model (LLM) to identify 1420 children's scientific drawings (covering 9 scientific themes/concepts), and uses the word2vec algorithm to calculate their semantic similarity. The study explores whether there are consistent drawing representations for children on the same theme, and attempts to establish a norm for children's scientific drawings, providing a baseline reference for follow-up children's drawing research. The results show that the representation of most drawings has consistency, manifested as most semantic similarity>0.8. At the same time, it was found that the consistency of the representation is independent of the accuracy (of LLM's recognition), indicating the existence of consistency bias. In the subsequent exploration of influencing factors, we used Kendall rank correlation coefficient to investigate the effects of "sample size", "abstract degree", and "focus points" on drawings, and used word frequency statistics to explore whether children represented abstract themes/concepts by reproducing what was taught in class. It was found that accuracy (of LLM's recognition) is the most sensitive indicator, and data such as sample size and semantic similarity are related to it; The consistency between classroom experiments and teaching purpose is also an important factor, many students focus more on the experiments themselves rather than what they explain.

Fan Wei, Yuan Li, Lior Shamir

In the past few years, computer vision and pattern recognition systems have been becoming increasingly more powerful, expanding the range of automatic tasks enabled by machine vision. Here we show that computer analysis of building images can perform quantitative analysis of architecture, and quantify similarities between city architectural styles in a quantitative fashion. Images of buildings from 18 cities and three countries were acquired using Google StreetView, and were used to train a machine vision system to automatically identify the location of the imaged building based on the image visual content. Experimental results show that the automatic computer analysis can automatically identify the geographical location of the StreetView image. More importantly, the algorithm was able to group the cities and countries and provide a phylogeny of the similarities between architectural styles as captured by StreetView images. These results demonstrate that computer vision and pattern recognition algorithms can perform the complex cognitive task of analyzing images of buildings, and can be used to measure and quantify visual similarities and differences between different styles of architectures. This experiment provides a new paradigm for studying architecture, based on a quantitative approach that can enhance the traditional manual observation and analysis. The source code used for the analysis is open and publicly available.

Changsheng Li, Junchi Yan, Fan Wei et al.

In this paper, we propose a novel multi-task learning (MTL) framework, called Self-Paced Multi-Task Learning (SPMTL). Different from previous works treating all tasks and instances equally when training, SPMTL attempts to jointly learn the tasks by taking into consideration the complexities of both tasks and instances. This is inspired by the cognitive process of human brain that often learns from the easy to the hard. We construct a compact SPMTL formulation by proposing a new task-oriented regularizer that can jointly prioritize the tasks and the instances. Thus it can be interpreted as a self-paced learner for MTL. A simple yet effective algorithm is designed for optimizing the proposed objective function. An error bound for a simplified formulation is also analyzed theoretically. Experimental results on toy and real-world datasets demonstrate the effectiveness of the proposed approach, compared to the state-of-the-art methods.

Changsheng Li, Fan Wei, Junchi Yan et al.

In this paper, we propose a novel multi-label learning framework, called Multi-Label Self-Paced Learning (MLSPL), in an attempt to incorporate the self-paced learning strategy into multi-label learning regime. In light of the benefits of adopting the easy-to-hard strategy proposed by self-paced learning, the devised MLSPL aims to learn multiple labels jointly by gradually including label learning tasks and instances into model training from the easy to the hard. We first introduce a self-paced function as a regularizer in the multi-label learning formulation, so as to simultaneously rank priorities of the label learning tasks and the instances in each learning iteration. Considering that different multi-label learning scenarios often need different self-paced schemes during optimization, we thus propose a general way to find the desired self-paced functions. Experimental results on three benchmark datasets suggest the state-of-the-art performance of our approach.

Changsheng Li, Fan Wei, Weishan Dong et al.

Online multiple-output regression is an important machine learning technique for modeling, predicting, and compressing multi-dimensional correlated data streams. In this paper, we propose a novel online multiple-output regression method, called MORES, for stream data. MORES can \emph{dynamically} learn the structure of the coefficients change in each update step to facilitate the model's continuous refinement. We observe that limited expressive ability of the regression model, especially in the preliminary stage of online update, often leads to the variables in the residual errors being dependent. In light of this point, MORES intends to \emph{dynamically} learn and leverage the structure of the residual errors to improve the prediction accuracy. Moreover, we define three statistical variables to \emph{exactly} represent all the seen samples for \emph{incrementally} calculating prediction loss in each online update round, which can avoid loading all the training data into memory for updating model, and also effectively prevent drastic fluctuation of the model in the presence of noise. Furthermore, we introduce a forgetting factor to set different weights on samples so as to track the data streams' evolving characteristics quickly from the latest samples. Experiments on one synthetic dataset and three real-world datasets validate the effectiveness of the proposed method. In addition, the update speed of MORES is at least 2000 samples processed per second on the three real-world datasets, more than 15 times faster than the state-of-the-art online learning algorithm.