Papers

Paste an arXiv link or drop any paper's PDF and the reading agents unroll it into a readable page: margin notes where the text gets hard, a guide for every figure, and an adversarial review of what the authors left unstated.

Ready to read

Douyin Multimodal Embedding Model Technical ReportMultimodal representation learning is a cornerstone of modern AI. By encoding multimodal queries and targets into vectors, it powers industrial applications such as search and recommendation, and increasingly underpins modern agents. Real-world platforms with complex modalities and massive-scale content, such as Douyin, Xiaohongshu, and YouTube, demand two capabilities simultaneously, efficiency under billion-scale indexing and fine-grained semantic discrimination for hard matching. Existing MLLM embedding models often struggle to jointly satisfy both requirements. Contrastive models are efficient but rely on pair-level supervision that is too coarse for fine-grained distinctions, while CoT-based models improve discrimination at the cost of explicit generation that is impractical to serve online. We present Douyin Multimodal Embedding (DME), a model that combines both strengths to meet the industrial demand for efficient and fine-grained representation. Specifically, DME is trained in two stages. Stage 1 performs large-scale contrastive pre-training that establishes a unified multimodal embedding space with broad modality and task coverage. Stage 2 supplements semantic sufficiency, the property that an embedding is grounded in retrieval-relevant evidence and preserves fine-grained counterpart-side semantics, through two complementary mechanisms. Evidence-Grounded Typed Latent Reasoning organizes retrieval evidence via hidden-space latent reasoning, and Cross-Conditional Reconstruction enforces counterpart-side semantics via cross-directional autoregressive reconstruction. Cross-Conditional Reconstruction is used only during training, and the latent tokens remain inside a single encoder forward pass and introduce only marginal query-side overhead. The generative supervision further makes DME embeddings information-complete, in that the input content can be recovered from them, which we quantify as an interpretable measure of semantic sufficiency and use to guide optimization in industrial settings. On MMEB-v2, DME achieves state-of-the-art results at comparable scales for both the 2B and 9B variants (74.8 and 78.4), with particularly strong performance on video and visual-document retrieval. In production, DME delivers a 2.92% relative improvement in overall score on Douyin’s in-house offline evaluation set, and has been deployed across a range of real-world Douyin scenarios such as generative search, image search, and AI search. Online A/B testing on Douyin search further confirms a 0.1% Lifetime (LT) gain.
WikiSkill: Compiling Agent Experience into Persistent Knowledge for Skill Evolution*Agent skills* package specialized knowledge and workflows into reusable resources that extend AI agent capabilities. Recent work automatically discovers such skills from agent experience, which enables agents to progressively adapt through interaction. However, the insights that guide skill development typically remain scattered across optimization histories, limiting their systematic reuse across iterations. We introduce WikiSkill, a framework that co-evolves agent skills with a *persistent* knowledge base (wiki). At a high level, WikiSkill separates raw execution experience, accumulated knowledge, and executable skills, while continuously consolidating experience into the wiki, which subsequent skill updates can build on. Across diverse benchmarks and models, WikiSkill consistently outperforms *state-of-the-art* skill-evolution methods and improves over no-skill baselines in most model-benchmark settings. We find that skill evolution complements model scaling: larger models generally benefit more from evolved skills, while smaller models with skills can outperform substantially larger models without them. We also find that evolved skills transfer effectively across models and model families, and skills evolved by other models can outperform self-evolved skills. Finally, our ablation studies confirm that persistent knowledge accumulation in the wiki is critical for effective skill evolution. These results demonstrate the benefits of systematically accumulating and refining agent experience for developing reusable and transferable skills.Liyan Tang, Cyrus Rashtchian, Chun-Sung Ferng, Andrew Tomkins, Da-Cheng Juan, Tu Vu
Restoring Voluntary Control of Locomotion after Paralyzing Spinal Cord InjuryHalf of human spinal cord injuries lead to chronic paralysis. Here, we introduce an electrochemical neuroprosthesis and a robotic postural interface designed to encourage supraspinally mediated movements in rats with paralyzing lesions. Despite the interruption of direct supraspinal pathways, the cortex regained the capacity to transform contextual information into task-specific commands to execute refined locomotion. This recovery relied on the extensive remodeling of cortical projections, including the formation of brainstem and intraspinal relays that restored qualitative control over electrochemically enabled lumbosacral circuitries. Automated treadmill-restricted training, which did not engage cortical neurons, failed to promote translesional plasticity and recovery. By encouraging active participation under functional states, our training paradigm triggered a cortex-dependent recovery that may improve function after similar injuries in humans.Rubia van den Brand, Janine Heutschi, Quentin Barraud, Jack DiGiovanna, Kay Bartholdi, Michèle Huerlimann, Lucia Friedli, Isabel Vollenweider, Eduardo Martin Moraud, Simone Duis, Nadia Dominici, Silvestro Micera, Pavel Musienko, Grégoire Courtine
Attention Is All You NeedThe dominant sequence transduction models are based on complex recurrent or convolutional neural networks that include an encoder and a decoder. The best performing models also connect the encoder and decoder through an attention mechanism. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. Experiments on two machine translation tasks show these models to be superior in quality while being more parallelizable and requiring significantly less time to train. Our model achieves 28.4 BLEU on the WMT 2014 English-to-German translation task, improving over the existing best results, including ensembles, by over 2 BLEU. On the WMT 2014 English-to-French translation task, our model establishes a new single-model state-of-the-art BLEU score of 41.8 after training for 3.5 days on eight GPUs, a small fraction of the training costs of the best models from the literature. We show that the Transformer generalizes well to other tasks by applying it successfully to English constituency parsing both with large and limited training data.Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin
Mixture-of-Agents Enhances Large Language Model CapabilitiesRecent advances in large language models (LLMs) demonstrate substantial capabilities in natural language understanding and generation tasks. With the growing number of LLMs, how to harness the collective expertise of multiple LLMs is an exciting open direction. Toward this goal, we propose a new approach that leverages the collective strengths of multiple LLMs through a Mixture-of-Agents (MoA) methodology. In our approach, we construct a layered MoA architecture wherein each layer comprises multiple LLM agents. Each agent takes all the outputs from agents in the previous layer as auxiliary information in generating its response. MoA models achieves state-of-art performance on AlpacaEval 2.0, MT-Bench and FLASK, surpassing GPT-4 Omni. For example, our MoA using only open-source LLMs is the leader of AlpacaEval 2.0 by a substantial gap, achieving a score of 65.1% compared to 57.5% by GPT-4 Omni.111Our code can be found in: <https://github.com/togethercomputer/moa>.Junlin Wang, Jue Wang, Ben Athiwaratkun, Ce Zhang, James Zou