Jialin Lu

Junyuan Zhang, Bin Wang, Qintong Zhang et al.

Table recognition (TR) aims to transform table images into semi-structured representations such as HTML or Markdown. As a core component of document parsing, TR has long relied on supervised learning, with recent efforts dominated by fine-tuning vision-language models (VLMs) using labeled data. While VLMs have brought TR to the next level, pushing performance further demands large-scale labeled data that is costly to obtain. Consequently, although proprietary models have continuously pushed the performance boundary, open-source models, often trained with limited resources and, in practice, the only viable option for many due to privacy regulations, still lag far behind. To bridge this gap, we introduce TRivia, a self-supervised fine-tuning method that enables pretrained VLMs to learn TR directly from unlabeled table images in the wild. Built upon Group Relative Policy Optimization, TRivia automatically identifies unlabeled samples that most effectively facilitate learning and eliminates the need for human annotations through a question-answering-based reward mechanism. An attention-guided module generates diverse questions for each table image, and the ability to interpret the recognition results and answer them correctly provides feedback to optimize the TR model. This closed-loop process allows the TR model to autonomously learn to recognize, structure, and reason over tables without labeled data. Leveraging this pipeline, we present TRivia-3B, an open-sourced, compact, and state-of-the-art TR model that surpasses existing systems (e.g., Gemini 2.5 Pro, MinerU2.5) on three popular benchmarks. Model and code are released at: https://github.com/opendatalab/TRivia

Junjie Shan, Ziqi Zhao, Jialin Lu et al.

Foundation models that bridge vision and language have made significant progress. While they have inspired many life-enriching applications, their potential for abuse in creating new threats remains largely unexplored. In this paper, we reveal that vision-language models (VLMs) can be weaponized to enhance gradient inversion attacks (GIAs) in federated learning (FL), where an FL server attempts to reconstruct private data samples from gradients shared by victim clients. Despite recent advances, existing GIAs struggle to reconstruct high-resolution images when the victim has a large local data batch. One promising direction is to focus reconstruction on valuable samples rather than the entire batch, but current methods lack the flexibility to target specific data of interest. To address this gap, we propose Geminio, the first approach to transform GIAs into semantically meaningful, targeted attacks. It enables a brand new privacy attack experience: attackers can describe, in natural language, the data they consider valuable, and Geminio will prioritize reconstruction to focus on those high-value samples. This is achieved by leveraging a pretrained VLM to guide the optimization of a malicious global model that, when shared with and optimized by a victim, retains only gradients of samples that match the attacker-specified query. Geminio can be launched at any FL round and has no impact on normal training (i.e., the FL server can steal clients' data while still producing a high-utility ML model as in benign scenarios). Extensive experiments demonstrate its effectiveness in pinpointing and reconstructing targeted samples, with high success rates across complex datasets and large batch sizes with resilience against defenses.

Jialin Lu, Martin Ester

Model interpretation, or explanation of a machine learning classifier, aims to extract generalizable knowledge from a trained classifier into a human-understandable format, for various purposes such as model assessment, debugging and trust. From a computaional viewpoint, it is formulated as approximating the target classifier using a simpler interpretable model, such as rule models like a decision set/list/tree. Often, this approximation is handled as standard supervised learning and the only difference is that the labels are provided by the target classifier instead of ground truth. This paradigm is particularly popular because there exists a variety of well-studied supervised algorithms for learning an interpretable classifier. However, we argue that this paradigm is suboptimal for it does not utilize the unique property of the model interpretation problem, that is, the ability to generate synthetic instances and query the target classifier for their labels. We call this the active-query property, suggesting that we should consider model interpretation from an active learning perspective. Following this insight, we argue that the active-query property should be employed when designing a model interpretation algorithm, and that the generation of synthetic instances should be integrated seamlessly with the algorithm that learns the model interpretation. In this paper, we demonstrate that by doing so, it is possible to achieve more faithful interpretation with simpler model complexity. As a technical contribution, we present an active algorithm Active Decision Set Induction (ADS) to learn a decision set, a set of if-else rules, for model interpretation. ADS performs a local search over the space of all decision sets. In every iteration, ADS computes confidence intervals for the value of the objective function of all local actions and utilizes active-query to determine the best one.