Multi-head latent attention (MLA) is designed to optimize KV cache memory through low-rank key-value joint compression. Rather than caching keys and values separately, MLA stores their compressed latent representations, reducing memory overhead while maintaining the performance. While MLA improves memory efficiency without compromising language model accuracy, its major limitation lies in its integration during the pre-training phase, requiring models to be trained from scratch. This raises a key question: can we use MLA's benefits fully or partially in models that have already been pre-trained with different attention mechanisms? In this paper, we propose X-EcoMLA to deploy post training distillation to enable the upcycling of Transformer-based attention into an efficient hybrid (i.e., combination of regular attention and MLA layers) or full MLA variant through lightweight post-training adaptation, bypassing the need for extensive pre-training. We demonstrate that leveraging the dark knowledge of a well-trained model can enhance training accuracy and enable extreme KV cache compression in MLA without compromising model performance. Our results show that using an 8B teacher model allows us to compress the KV cache size of the Llama3.2-1B-Inst baseline by 6.4x while preserving 100% of its average score across multiple tasks on the LM Harness Evaluation benchmark. This is achieved with only 3.6B training tokens and about 70 GPU hours on AMD MI300 GPUs, compared to the 370K GPU hours required for pre-training the Llama3.2-1B model.

The problem of relevant and diverse subset selection has a wide range of applications, including recommender systems and retrieval-augmented generation (RAG). For example, in recommender systems, one is interested in selecting relevant items, while providing a diversified recommendation. Constrained subset selection problem is NP-hard, and popular approaches such as Maximum Marginal Relevance (MMR) are based on greedy selection. Many real-world applications involve large data, but the original MMR work did not consider distributed selection. This limitation was later addressed by a method called DGDS which allows for a distributed setting using random data partitioning. Here, we exploit structure in the data to further improve both scalability and performance on the target application. We propose MUSS, a novel method that uses a multilevel approach to relevant and diverse selection. We provide a rigorous theoretical analysis and show that our method achieves a constant factor approximation of the optimal objective. In a recommender system application, our method can achieve the same level of performance as baselines, but 4.5 to 20 times faster. Our method is also capable of outperforming baselines by up to 6 percent points of RAG-based question answering accuracy.

Graph learning plays a vital role in mining and analyzing complex relationships involved in graph data, which is widely used in many real-world applications like transaction networks and communication networks. Foundation models in CV and NLP have shown powerful cross-domain capabilities that are also significant in graph domains. However, existing graph learning approaches struggle with cross-domain tasks. Inspired by successes in CV and NLP, cross-domain graph learning has once again become a focal point of attention to realizing true graph foundation models. In this survey, we present a comprehensive review and analysis of existing works on cross-domain graph learning. Concretely, we first propose a new taxonomy, categorizing existing approaches based on the learned cross-domain information: structure, feature, and structure-feature mixture. Next, we systematically survey representative methods in these categories. Finally, we discuss the remaining limitations of existing studies and highlight promising avenues for future research. Relevant papers are summarized and will be consistently updated at: https://github.com/cshhzhao/Awesome-Cross-Domain-Graph-Learning.

Large language models (LLMs) have evolved beyond simple text generation to power software agents that directly translate natural language commands into tangible actions. While API-based LLM agents initially rose to prominence for their robust automation capabilities and seamless integration with programmatic endpoints, recent progress in multimodal LLM research has enabled GUI-based LLM agents that interact with graphical user interfaces in a human-like manner. Although these two paradigms share the goal of enabling LLM-driven task automation, they diverge significantly in architectural complexity, development workflows, and user interaction models. This paper presents the first comprehensive comparative study of API-based and GUI-based LLM agents, systematically analyzing their divergence and potential convergence. We examine key dimensions and highlight scenarios in which hybrid approaches can harness their complementary strengths. By proposing clear decision criteria and illustrating practical use cases, we aim to guide practitioners and researchers in selecting, combining, or transitioning between these paradigms. Ultimately, we indicate that continuing innovations in LLM-based automation are poised to blur the lines between API- and GUI-driven agents, paving the way for more flexible, adaptive solutions in a wide range of real-world applications.

Vision-Language Models have made significant progress on many perception-focused tasks. However, their progress on reasoning-focused tasks remains limited due to the lack of high-quality and diverse training data. In this work, we aim to address the scarcity of reasoning-focused multimodal datasets. We propose VisualWebInstruct, a novel approach that leverages search engines to create a diverse and high-quality dataset spanning multiple disciplines, including mathematics, physics, finance, and chemistry, etc. Starting with a meticulously selected set of 30,000 seed images, we employ Google Image Search to identify websites containing similar images. We collect and process HTML data from over 700K unique URLs. Through a pipeline of content extraction, filtering, and synthesis, we construct a dataset of approximately 900K question-answer (QA) pairs, with 40% consisting of visual QA pairs and the remaining comprising text-based QA pairs. Models fine-tuned on VisualWebInstruct demonstrate significant performance improvements: (1) fine-tuning on Llava-OV results in 10-20 absolute points improvement across benchmarks, and (2) fine-tuning from MAmmoTH-VL yields a 5 absolute points gain across benchmarks. Our best model, MAmmoTH-VL2, achieves state-of-the-art performance within the 10B parameter class on MMMU-Pro (40.7), MathVerse (42.6), and DynaMath (55.7). These results highlight the effectiveness of our dataset in enhancing the reasoning capabilities of vision-language models for complex multimodal tasks.

In personalized technology and psychological research, precisely detecting demographic features and personality traits from digital interactions becomes ever more important. This work investigates implicit categorization, inferring personality and gender variables directly from linguistic patterns in Telegram conversation data, while conventional personality prediction techniques mostly depend on explicitly self-reported labels. We refine a Transformer-based language model (RoBERTa) to capture complex linguistic cues indicative of personality traits and gender differences using a dataset comprising 138,866 messages from 1,602 users annotated with MBTI types and 195,016 messages from 2,598 users annotated with gender. Confidence levels help to greatly raise model accuracy to 86.16\%, hence proving RoBERTa's capacity to consistently identify implicit personality types from conversational text data. Our results highlight the usefulness of Transformer topologies for implicit personality and gender classification, hence stressing their efficiency and stressing important trade-offs between accuracy and coverage in realistic conversational environments. With regard to gender classification, the model obtained an accuracy of 74.4\%, therefore capturing gender-specific language patterns. Personality dimension analysis showed that people with introverted and intuitive preferences are especially more active in text-based interactions. This study emphasizes practical issues in balancing accuracy and data coverage as Transformer-based models show their efficiency in implicit personality and gender prediction tasks from conversational texts.

Multimodal retrieval-augmented generation (RAG) enhances the visual reasoning capability of vision-language models (VLMs) by dynamically accessing information from external knowledge bases. In this work, we introduce \textit{Poisoned-MRAG}, the first knowledge poisoning attack on multimodal RAG systems. Poisoned-MRAG injects a few carefully crafted image-text pairs into the multimodal knowledge database, manipulating VLMs to generate the attacker-desired response to a target query. Specifically, we formalize the attack as an optimization problem and propose two cross-modal attack strategies, dirty-label and clean-label, tailored to the attacker's knowledge and goals. Our extensive experiments across multiple knowledge databases and VLMs show that Poisoned-MRAG outperforms existing methods, achieving up to 98\% attack success rate with just five malicious image-text pairs injected into the InfoSeek database (481,782 pairs). Additionally, We evaluate 4 different defense strategies, including paraphrasing, duplicate removal, structure-driven mitigation, and purification, demonstrating their limited effectiveness and trade-offs against Poisoned-MRAG. Our results highlight the effectiveness and scalability of Poisoned-MRAG, underscoring its potential as a significant threat to multimodal RAG systems.

This paper argues that a dataset's legal risk cannot be accurately assessed by its license terms alone; instead, tracking dataset redistribution and its full lifecycle is essential. However, this process is too complex for legal experts to handle manually at scale. Tracking dataset provenance, verifying redistribution rights, and assessing evolving legal risks across multiple stages require a level of precision and efficiency that exceeds human capabilities. Addressing this challenge effectively demands AI agents that can systematically trace dataset redistribution, analyze compliance, and identify legal risks. We develop an automated data compliance system called NEXUS and show that AI can perform these tasks with higher accuracy, efficiency, and cost-effectiveness than human experts. Our massive legal analysis of 17,429 unique entities and 8,072 license terms using this approach reveals the discrepancies in legal rights between the original datasets before redistribution and their redistributed subsets, underscoring the necessity of the data lifecycle-aware compliance. For instance, we find that out of 2,852 datasets with commercially viable individual license terms, only 605 (21%) are legally permissible for commercialization. This work sets a new standard for AI data governance, advocating for a framework that systematically examines the entire lifecycle of dataset redistribution to ensure transparent, legal, and responsible dataset management.

Continual learning has become a promising solution to refine large language models incrementally by leveraging user feedback. In particular, online continual learning - iteratively training the model with small batches of user feedback - has demonstrated notable performance improvements. However, the existing practice of separating training and serving processes forces the online trainer to recompute the intermediate results already done during serving. Such redundant computations can account for 30%-42% of total training time. In this paper, we propose Alchemist, to the best of our knowledge, the first online continual learning system that efficiently reuses serving activations to increase training throughput. Alchemist introduces two key techniques: (1) recording and storing activations and KV cache only during the prefill phase to minimize latency and memory overhead; and (2) smart activation offloading and hedging. Evaluations with inputs of varied token length sampled from ShareGPT dataset show that compared with a separate training cluster, Alchemist significantly increases training throughput by up to 1.72x, reduces up to 47% memory usage during training, and supports up to 2x more training tokens - all while maintaining negligible impact on serving latency.

However, human auditory perception is inherently selective: listeners focus on specific speakers while ignoring others in complex auditory scenes. Existing models do not incorporate this selectivity, limiting their ability to generate perceptionaligned responses. To address this, we introduce Intention-Informed Auditory Scene Understanding (II-ASU) and present Auditory Attention-Driven LLM (AAD-LLM), a prototype system that integrates brain signals to infer listener attention. AAD-LLM extends an auditory LLM by incorporating intracranial electroencephalography (iEEG) recordings to decode which speaker a listener is attending to and refine responses accordingly. The model first predicts the attended speaker from neural activity, then conditions response generation on this inferred attentional state. We evaluate AAD-LLM on speaker description, speech transcription and extraction, and question answering Figure 1: AAD-LLM is a brain-computer interface in multitalker scenarios, with both objective (BCI) for auditory scene understanding. It decodes neural and subjective ratings showing improved alignment signals to identify the attended speaker and integrates with listener intention. By taking a first this information into a language model, generating responses step toward intention-aware auditory AI, this that align with the listener's perceptual focus.