Automated scoring plays a crucial role in education by reducing the reliance on human raters and offering scalable and immediate evaluation of student work. While large language models (LLMs) have shown strong potential in this task, their use as end-to-end raters faces challenges such as low accuracy, prompt sensitivity, limited interpretability, and rubric misalignment, which hinder practical implementation. To address the limitations, we propose AutoSCORE, a multi-agent LLM framework enhancing automated scoring via rubric-aligned Structured COmponent REcognition. With two agents, AutoSCORE first extracts rubric-relevant components from student responses and encodes them into a structured representation (i.e., Scoring Rubric Component Extraction Agent), which is then used to assign final scores (i.e., Scoring Agent). This design ensures that model reasoning follows a human-like grading process, enhancing interpretability and robustness. We evaluate AutoSCORE on four benchmark datasets from the ASAP benchmark, using both proprietary and open-source LLMs (GPT-4o, LLaMA-3.1-8B, LLaMA-3.1-70B). Across diverse tasks and rubrics, AutoSCORE predominantly improves scoring accuracy, human-machine agreement (QWK, correlations), and reduces error metrics (MAE, RMSE) compared to single-agent baselines, with particularly strong benefits on complex, multidimensional rubrics, and especially large relative gains on smaller LLMs. These results demonstrate that structured component recognition combined with multi-agent design offers a scalable, reliable, and interpretable solution for automated scoring.

The rising prevalence of vision-threatening eye diseases poses a major global health and economic burden, yet timely diagnosis remains limited by workforce shortages, diagnostic delays, and restricted access to specialized care. Artificial intelligence (AI) offers potential solutions. In particular, recent progress in foundation models and large language models-especially multimodal large language models (MLLMs)-has shown promise in medical image interpretation and automated clinical documentation. However, advancing MLLMs for ophthalmology is hindered by the lack of unified, comprehensive benchmark datasets for development and evaluation. Most existing benchmarks were designed for earlier models, which focused on narrow tasks or specific disease conditions. These benchmarks typically provide outputs in the form of disease labels rather than free-text responses. As a result, they are less suitable for assessing emerging generative models. In this work, we present LMOD+, a large-scale multimodal ophthalmology benchmark dataset comprising 32,633 instances with multi-granular annotations across 12 common ophthalmic conditions and 5 imaging modalities. The dataset integrates imaging, anatomical structures, demographics, and free-text annotations. It supports primary ophthalmic applications such as anatomical structure recognition, disease screening, disease staging, and demographic prediction for potential performance bias evaluation. Alongside the dataset, we introduce a systematic and unified data curation pipeline that repurposes existing or new datasets for MLLM development. LMOD+ extends our preliminary LMOD benchmark-the first multimodal ophthalmology benchmark for MLLMs-with three major enhancements. First, we expanded the dataset by nearly 50% (from 21,933 to 32,633 instances). The color fundus photography (CFP) modality, the most accessible imaging modality in ophthalmology, was significantly enlarged to cover a broader range of pathological conditions. Second, we broadened task coverage to include (a) 12 binary disease diagnosis tasks for prevalent conditions such as diabetic retinopathy, age-related macular degeneration, and retinal vein occlusion; (b) multi-class ophthalmic disease diagnosis; (c) disease severity classification, including a diabetic retinopathy staging task, which uses two internationally adopted grading standards: the international clinical diabetic retinopathy classification and the Scottish diabetic retinopathy grading scheme classification; and (d) demographic prediction (age and sex) to assess potential model bias. Third, we systematically evaluated 24 state-of-the-art MLLMs, including recent models from the InternVL, Qwen, and DeepSeek families. Our evaluations highlight both the promise and limitations of current MLLMs in ophthalmology. For example, Qwen-7B and InternVL achieved accuracies of 58.26% and 57.83% in disease screening under a zero-shot setting with a single model-a considerably more challenging paradigm than traditional fine-tuning, where separate models are trained for each specific task. InternVL also demonstrated potential in anatomical recognition. Nonetheless, overall performance remained suboptimal and often close to random baselines for challenging tasks such as disease staging, underscoring the substantial gap between general-domain MLLMs and the specialized requirements of ophthalmology. We publicly release the dataset, curation pipeline, and leaderboard to encourage community-wide development and evaluation of MLLMs, with the goal of advancing ophthalmic applications and ultimately reducing the global burden of vision-threatening diseases through AI. The dataset website, benchmark leaderboard, and download link are available at https://kfzyqin.github.io/lmod_plus.

Explainable AI (XAI) refers to techniques that provide human-understandable insights into the workings of AI models. Recently, the focus of XAI has been extended toward explaining Large Language Models (LLMs). This extension calls for a significant transformation in the XAI methodologies for two reasons. First, many existing XAI methods cannot be directly applied to LLMs due to their complexity and advanced capabilities. Second, as LLMs are increasingly deployed in diverse applications, the role of XAI shifts from merely opening the “black box” to actively enhancing the productivity and applicability of LLMs in real-world settings. Meanwhile, the conversation and generation abilities of LLMs can reciprocally enhance XAI. Therefore, in this article, we introduce Usable XAI in the context of LLMs by analyzing (1) how XAI can explain and improve LLM-based AI systems and (2) how XAI techniques can be improved by using LLMs. We introduce 10 strategies, introducing the key techniques for each and discussing their associated challenges. We also provide case studies to demonstrate how to obtain and leverage explanations.

Sparse Autoencoders (SAEs) have recently emerged as powerful tools for interpreting and steering the internal representations of large language models (LLMs).However, conventional approaches to analyzing SAEs typically rely solely on input-side activations, without considering the causal influence between each latent feature and the model's output.This work is built on two key hypotheses: (1) activated latents do not contribute equally to the construction of the model's output, and (2) only latents with high causal influence are effective for model steering.To validate these hypotheses, we propose Gradient Sparse Autoencoder (GradSAE), a simple yet effective method that identifies the most influential latents by incorporating output-side gradient information.Our code is available at https: //github.com/Tizzzzy/sae_gradient.

Vector-Quantized Generative Models (VQGMs) have emerged as powerful tools for image generation. However, the key component of VQGMs—the codebook of discrete tokens—is still not well understood, e.g., which tokens are critical to generate an image of a certain concept? This paper introduces Concept-Oriented Token Explanation (CORTEX), a novel approach for interpreting VQGMs by identifying concept-specific token combinations. Our framework employs two methods: (1) a sample-level explanation method that analyzes token importance scores in individual images, and (2) a codebook-level explanation method that explores the entire codebook to find globally relevant tokens. Experimental results demonstrate CORTEX’s efficacy in providing clear explanations of token usage in the generative process, outperforming baselines across multiple pretrained VQGMs. Besides enhancing VQGMs transparency, CORTEX is useful in applications such as targeted image editing and shortcut feature detection. Our code is available at https://github.com/YangTianze009/CORTEX.

Modern text classification methods heavily rely on contextual embeddings from large language models (LLMs). Compared to human-engineered features, these embeddings provide automatic and effective representations for classification model training. However, they also introduce a challenge: we lose the ability to manually remove unintended features, such as sensitive or task-irrelevant features, to guarantee regulatory compliance or improve the generalizability of classification models. This limitation arises because LLM embeddings are opaque and difficult to interpret. In this paper, we propose a novel framework to identify and regularize unintended features in the LLM latent space. Specifically, we first pre-train a sparse autoencoder (SAE) to extract interpretable features from LLM latent spaces. To ensure the SAE can capture task-specific features, we further fine-tune it on task-specific datasets. In training the classification model, we propose a simple and effective regularizer, by minimizing the similarity between the classifier weights and the identified unintended feature, to remove the impact of these unintended features on classification. We evaluate the proposed framework on three real-world tasks, including toxic chat detection, reward modeling, and disease diagnosis. Results show that the proposed self-regularization framework can improve the classifier's generalizability by regularizing those features that are not semantically correlated to the task. This work pioneers controllable text classification on LLM latent spaces by leveraging interpreted features to address generalizability, fairness, and privacy challenges. The code and data are publicly available at https://github.com/JacksonWuxs/Controllable_LLM_Classifier.

Large Language Models (LLMs) have shown proficiency in question-answering tasks but often struggle to integrate real-time knowledge, leading to potentially outdated or inaccurate responses. This problem becomes even more challenging when dealing with multi-hop questions, since they require LLMs to update and integrate multiple knowledge pieces relevant to the questions. To tackle the problem, we propose the Retrieval-Augmented model Editing (RAE) framework for multi-hop question answering. RAE first retrieves edited facts and then refines the language model through in-context learning. Specifically, our retrieval approach, based on mutual information maximization, leverages the reasoning abilities of LLMs to identify chain facts that traditional similarity-based searches might miss. In addition, our framework includes a pruning strategy to eliminate redundant information from the retrieved facts, which enhances the editing accuracy and mitigates the hallucination problem. Our framework is supported by theoretical justification for its fact retrieval efficacy. Finally, comprehensive evaluation across various LLMs validates RAE's ability in providing accurate answers with updated knowledge. Our code is available at: https://github.com/sycny/RAE.

Xuansheng Wu, Wenlin Yao, Jianshu Chen, Xiaoman Pan, Xiaoyang Wang, Ninghao Liu, Dong Yu. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Recommending products to users with intuitive explanations helps improve the system in transparency, persuasiveness, and satisfaction. Existing interpretation techniques include post hoc methods and interpretable modeling. The former category could quantitatively analyze input contribution to model prediction but has limited interpretation faithfulness, while the latter could explain model internal mechanisms but may not directly attribute model predictions to input features. In this study, we propose a novel Dual Interpretable Recommendation model called DIRECT, which integrates ideas of the two interpretation categories to inherit their advantages and avoid limitations. Specifically, DIRECT makes use of item descriptions as explainable evidence for recommendation. First, similar to the post hoc interpretation, DIRECT could attribute the prediction of a user preference score to textual words of the item descriptions. The attribution of each word is related to its sentiment polarity and word importance, where a word is important if it corresponds to an item aspect that the user is interested in. Second, to improve the interpretability of embedding space, we propose to extract high-level concepts from embeddings, where each concept corresponds to an item aspect. To learn discriminative concepts, we employ a concept bottleneck layer and maximize the coding rate reduction on word-aspect embeddings by leveraging a word–word affinity graph extracted from a pre-trained language model. In this way, DIRECT simultaneously achieves faithful attribution and usable interpretation of embedding space. We also show that DIRECT achieves linear inference time complexity regarding the length of item reviews. We conduct experiments including ablation studies on five real-world datasets. Quantitative analysis, visualizations, and case studies verify the interpretability of DIRECT. Our code is available at: https://github.com/JacksonWuxs/DIRECT .

Backdoor attacks inject poisoned samples into the training data, resulting in the misclassification of the poisoned input during a model's deployment. Defending against such attacks is challenging, especially for real-world black-box models where only query access is permitted. In this paper, we propose a novel defense framework against backdoor attacks through Zero-shot Image Purification (ZIP). Our framework can be applied to poisoned models without requiring internal information about the model or any prior knowledge of the clean/poisoned samples. Our defense framework involves two steps. First, we apply a linear transformation (e.g., blurring) on the poisoned image to destroy the backdoor pattern. Then, we use a pre-trained diffusion model to recover the missing semantic information removed by the transformation. In particular, we design a new reverse process by using the transformed image to guide the generation of high-fidelity purified images, which works in zero-shot settings. We evaluate our ZIP framework on multiple datasets with different types of attacks. Experimental results demonstrate the superiority of our ZIP framework compared to state-of-the-art backdoor defense baselines. We believe that our results will provide valuable insights for future defense methods for black-box models. Our code is available at https://github.com/sycny/ZIP.

While Transformers have yielded impressive results for video classification on large datasets recently, simpler models without the transformer architecture can be promising for small datasets. In this paper, we propose three major techniques to improve feature quality and another three to alleviate overfitting in an attempt to make lightweight models achieve higher performances. In particular, we enhance features of Image Flow by combining temporal information, multi-level features of CNNs, and Text embedding. We alleviate overfitting by removing redundant modal, fine-tuning dropout rate, and augmenting data. In the 2021 Tencent Advertisement Algorithm Competition, the baseline model achieved a GAP score of 0.8019 offline with our strategies. It's worth mentioning that our design works well with the 10-fold method, which produces our final submitting model with a GAP score of 0.8210 online, ranking the 5th among 287 teams. In addition, our solution is among the fastest within the top 10 teams.