Language model inversion (LMI), i.e., recovering hidden prompts from outputs, emerges as a concrete threat to user privacy and system security. We recast LMI as reusing the LLM's own latent space and propose the Invariant Latent Space Hypothesis (ILSH): (1) diverse outputs from the same source prompt should preserve consistent semantics (source invariance), and (2) inputoutput cyclic mappings should be self-consistent within a shared latent space (cyclic invariance). Accordingly, we present Inv2A, which treats the LLM as an invariant decoder and learns only a lightweight inverse encoder that maps outputs to a denoised pseudo-representation. When multiple outputs are available, they are sparsely concatenated at the representation layer to increase information density. Training proceeds in two stages: contrastive alignment (source invariance) and supervised reinforcement (cyclic invariance). An optional training-free neighborhood search can refine local performance. Across 9 datasets covering user and system prompt scenarios, Inv2A outperforms baselines by an average of 4.77% BLEU score while reducing dependence on large inverse corpora. Our analysis further shows that prevalent defenses provide limited protection, underscoring the need for stronger strategies.

While Large Language Models (LLMs) excel at many reasoning tasks, their native inability to produce calibrated, multi-class probability distributions limits their use in high-stakes Web applications like content moderation and fraud detection. Existing methods to elicit probabilities from LLMs either sacrifice their crucial Chain-of-Thought (CoT) reasoning capabilities or suffer from poor calibration. To address this, we introduce a new paradigm, Verbalized Probability Distribution, and a novel training framework, RLVP (Reinforcement Learning with Verbalized Probabilities). RLVP fine-tunes an LLM to generate both an interpretable CoT and a complete, verbalized probability distribution. We overcome the ''insufficient reward granularity'' problem in standard Reinforcement Learning (RL) for classification by using soft probabilities from expert tabular models as a dense reward curriculum. Through large-scale joint training on 169 tabular tasks, we demonstrate that a single RLVP-trained model can surpass a strong, task-specific XGBoost baseline on up to 55% of tasks. More importantly, the trained model achieves state-of-the-art few-shot performance on unseen, heterogeneous Web benchmarks that mix structured data with free text, achieving performance comparable to or superior than expert models trained on the same limited data. This showcases a strong capability for generalization and knowledge transfer to complex Web data. Our work presents a viable path toward building general-purpose, probabilistically-sound, and interpretable foundation models for the Web.

Zhanming Shen, Hao Chen, Yulei Tang, Shaolin Zhu, Wentao Ye, Xiaomeng Hu, Haobo Wang, Gang Chen, Junbo Zhao. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Qi Zhang, Shouqing Yang, Lirong Gao, Hao Chen, Xiaomeng Hu, Jinglei Chen, Jiexiang Wang, Sheng Guo, Bo Zheng, Haobo Wang, Junbo Zhao. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

The Electroencephalogram (EEG)-based emotion recognition is promising yet limited by the requirement of a large number of training data. Collecting substantial labeled samples in the training trials is the key to the generalization on the test trials. This process is time-consuming and laborious. In recent years, several studies have proposed various semisupervised learning (e.g., active learning) and transfer learning (e.g., domain adaptation, style transfer mapping) methods to alleviate the requirement on training data. However, most of them are iterative methods, which need considerable training time and are unfeasible in practice. To tackle this problem, we present the Fast Online Instance Transfer (FOIT) for improved affective Brain-computer Interface (aBCI). FOIT selects auxiliary data from historical sessions and (or) other subjects heuristically, which are then combined with the training data for supervised training. The predictions on the test trials are made by an ensemble classifier. As a one-shot algorithm, FOIT avoids the time-consuming iterations. Experimental results show that FOIT brings significant improvement in accuracy for the three-category classification (1%-8%) on the SEED dataset and four-category classification (1%-14%) on the SEED-IV dataset in the cross-subject, cross-session and cross-all scenarios. The time cost over the baselines is moderate (~35s on average for our machine). In contrast, to achieve comparative accuracies, the iterative methods require much more time (~45s - ~900s). FOIT provides a simple, fast and practically feasible solution to improve the generalization of aBCIs and allows various choices of classifiers without constraints. Our codes are available online.