Training large Mixture-of-Experts (MoE) models remains computationally prohibitive due to their extreme compute and memory demands. Although low-precision training promises to accelerate computation and reduce memory footprint, existing implementations still rely on BF16-dominated dataflows with frequent quantize-dequantize (Q/DQ) conversions. These redundant casts erode much of FP8's theoretical efficiency. However, naively removing these casts by keeping dataflows entirely in FP8 introduces double quantization error: tensors quantized along different dimensions accumulate inconsistent scaling factors, degrading numerical stability. We propose FP8-Flow-MoE, an FP8 training recipe featuring a quantization-consistent FP8-centric dataflow with a scaling-aware transpose and fused FP8 operators that streamline computation and eliminate explicit cast operations from 12 to 2. Evaluations on a 671B-parameter MoE model demonstrate up to 21\% higher throughput and 16.5 GB lower memory usage per GPU compared to BF16 and naïve FP8 baselines, while maintaining stable convergence. We provide a plug-and-play FP8 recipe compatible with TransformerEngine and Megatron-LM, which will be open-sourced soon.

We propose a Graph Neural Network (GNN)-based approach for Handwritten Mathematical Expression (HME) recognition by modeling HMEs as graphs, where nodes represent symbols and edges capture spatial dependencies. A deep BLSTM network is used for symbol segmentation, recognition, and spatial relation classification, forming an initial primitive graph. A 2D-CFG parser then generates all possible spatial relations, while the GNN-based link prediction model refines the structure by removing unnecessary connections, ultimately forming the Symbol Label Graph. Experimental results demonstrate the effectiveness of our approach, showing promising performance in HME structure recognition.

We introduce SAIL-RL, a reinforcement learning (RL) post-training framework that enhances the reasoning capabilities of multimodal large language models (MLLMs) by teaching them when and how to think. Existing approaches are limited by outcome-only supervision, which rewards correct answers without ensuring sound reasoning, and by uniform thinking strategies, which often lead to overthinking on simple tasks and underthinking on complex ones. SAIL-RL addresses these challenges with a dual reward system: the Thinking Reward, which evaluates reasoning quality through factual grounding, logical coherence, and answer consistency, and the Judging Reward, which adaptively determines whether deep reasoning or direct answering is appropriate. Experiments on the state-of-the-art SAIL-VL2 show that SAIL-RL improves reasoning and multimodal understanding benchmarks at both 4B and 8B scales, achieving competitive performance against commercial closed-source models such as GPT-4o, and substantially reduces hallucinations, establishing it as a principled framework for building more reliable and adaptive MLLMs. The code will be available at https://github.com/BytedanceDouyinContent/SAIL-RL.

Serving large generative models such as LLMs and multi- modal transformers requires balancing user-facing SLOs (e.g., time-to-first-token, time-between-tokens) with provider goals of efficiency and cost reduction. Existing solutions rely on static provisioning or model-level autoscaling, both of which treat the model as a monolith. This coarse-grained resource management leads to degraded performance or significant resource underutilization due to poor adaptability to dynamic inference traffic that is common online. The root cause of this inefficiency lies in the internal structure of generative models: they are executed as graphs of interconnected operators. Through detailed characterization and systematic analysis, we find that operators are heterogeneous in their compute and memory footprints and exhibit diverse sensitivity to workload and resource factors such as batch size, sequence length, and traffic rate. This heterogeneity suggests that the operator, rather than the entire model, is the right granularity for scaling decisions. We propose an operator-level autoscaling framework, which allocates resources at finer (operator)-granularity, optimizing the scaling, batching, and placement based on individual operator profiles. Evaluated on production-scale traces, our approach preserves SLOs with up to 40% fewer GPUs and 35% less energy, or under fixed resources achieves 1.6x higher throughput with 5% less energy. These results show that the operator, rather than the model, is fundamentally a more effective unit for scaling large generative workloads.

Novel research ideas play a critical role in advancing scientific inquiries. Recent advancements in Large Language Models (LLMs) have demonstrated their potential to generate novel research ideas by leveraging large-scale scientific literature. However, previous work in research ideation has primarily relied on simplistic methods, such as keyword co-occurrence or semantic similarity. These approaches focus on identifying statistical associations in the literature but overlook the complex, contextual relationships between scientific concepts, which are essential to effectively leverage knowledge embedded in human literature. For instance, papers that simultaneously mention "keyword A" and "keyword B" often present research ideas that integrate both concepts. Additionally, some LLM-driven methods propose and refine research ideas using the model's internal knowledge, but they fail to effectively utilize the scientific concept network, limiting the grounding of ideas in established research. To address these challenges, we propose the Deep Ideation framework to address these challenges, integrating a scientific network that captures keyword co-occurrence and contextual relationships, enriching LLM-driven ideation. The framework introduces an explore-expand-evolve workflow to iteratively refine research ideas, using an Idea Stack to track progress. A critic engine, trained on real-world reviewer feedback, guides the process by providing continuous feedback on the novelty and feasibility of ideas. Our experiments show that our approach improves the quality of generated ideas by 10.67% compared to other methods, with ideas surpassing top conference acceptance levels. Human evaluation highlights their practical value in scientific research, and ablation studies confirm the effectiveness of each component in the workflow. Code repo is available at https://github.com/kyZhao-1/Deep-Ideation.

Alzheimer's disease (AD) is the most prevalent form of dementia, and its early diagnosis is essential for slowing disease progression. Recent studies on multimodal neuroimaging fusion using MRI and PET have achieved promising results by integrating multi-scale complementary features. However, most existing approaches primarily emphasize cross-modal complementarity while overlooking the diagnostic importance of modality-specific features. In addition, the inherent distributional differences between modalities often lead to biased and noisy representations, degrading classification performance. To address these challenges, we propose a Collaborative Attention and Consistent-Guided Fusion framework for MRI and PET based AD diagnosis. The proposed model introduces a learnable parameter representation (LPR) block to compensate for missing modality information, followed by a shared encoder and modality-independent encoders to preserve both shared and specific representations. Furthermore, a consistency-guided mechanism is employed to explicitly align the latent distributions across modalities. Experimental results on the ADNI dataset demonstrate that our method achieves superior diagnostic performance compared with existing fusion strategies.

Agents that understand objects and their interactions can learn policies that are more robust and transferable. However, most object-centric RL methods factor state by individual objects while leaving interactions implicit. We introduce the Factored Interactive Object-Centric World Model (FIOC-WM), a unified framework that learns structured representations of both objects and their interactions within a world model. FIOC-WM captures environment dynamics with disentangled and modular representations of object interactions, improving sample efficiency and generalization for policy learning. Concretely, FIOC-WM first learns object-centric latents and an interaction structure directly from pixels, leveraging pre-trained vision encoders. The learned world model then decomposes tasks into composable interaction primitives, and a hierarchical policy is trained on top: a high level selects the type and order of interactions, while a low level executes them. On simulated robotic and embodied-AI benchmarks, FIOC-WM improves policy-learning sample efficiency and generalization over world-model baselines, indicating that explicit, modular interaction learning is crucial for robust control.

With the growing computational demands of large language models (LLMs), efficient inference has become increasingly critical for practical deployment. Depth pruning has emerged as a promising approach for reducing the computational costs of large language models by removing transformer layers. However, existing methods typically rely on fixed block masks, which can lead to suboptimal performance across different tasks and inputs. In this paper, we propose IG-Pruning, a novel input-aware block-wise pruning method that dynamically selects layer masks at inference time. Our approach consists of two stages: (1) Discovering diverse mask candidates through semantic clustering and L0 optimization, and (2) Implementing efficient dynamic pruning without the need for extensive training. Experimental results demonstrate that our method consistently outperforms state-of-the-art static depth pruning methods, making it particularly suitable for resource-constrained deployment scenarios.

Segmental longitudinal strain (SLS) of the left ventricle (LV) is an important prognostic indicator for evaluating regional LV dysfunction, in particular for diagnosing and managing myocardial ischemia. Current techniques for strain estimation require significant manual intervention and expertise, limiting their efficiency and making them too resource-intensive for monitoring purposes. This study introduces the first automated pipeline, autoStrain, for SLS estimation in transesophageal echocardiography (TEE) using deep learning (DL) methods for motion estimation. We present a comparative analysis of two DL approaches: TeeFlow, based on the RAFT optical flow model for dense frame-to-frame predictions, and TeeTracker, based on the CoTracker point trajectory model for sparse long-sequence predictions. As ground truth motion data from real echocardiographic sequences are hardly accessible, we took advantage of a unique simulation pipeline (SIMUS) to generate a highly realistic synthetic TEE (synTEE) dataset of 80 patients with ground truth myocardial motion to train and evaluate both models. Our evaluation shows that TeeTracker outperforms TeeFlow in accuracy, achieving a mean distance error in motion estimation of 0.65 mm on a synTEE test dataset. Clinical validation on 16 patients further demonstrated that SLS estimation with our autoStrain pipeline aligned with clinical references, achieving a mean difference (95\% limits of agreement) of 1.09% (-8.90% to 11.09%). Incorporation of simulated ischemia in the synTEE data improved the accuracy of the models in quantifying abnormal deformation. Our findings indicate that integrating AI-driven motion estimation with TEE can significantly enhance the precision and efficiency of cardiac function assessment in clinical settings.

Strawberries are among the most economically significant fruits in the United States, generating over $2 billion in annual farm-gate sales and accounting for approximately 13% of the total fruit production value. Plant phenotyping plays a vital role in selecting superior cultivars by characterizing plant traits such as morphology, canopy structure, and growth dynamics. However, traditional plant phenotyping methods are time-consuming, labor-intensive, and often destructive. Recently, neural rendering techniques, notably Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS), have emerged as powerful frameworks for high-fidelity 3D reconstruction. By capturing a sequence of multi-view images or videos around a target plant, these methods enable non-destructive reconstruction of complex plant architectures. Despite their promise, most current applications of 3DGS in agricultural domains reconstruct the entire scene, including background elements, which introduces noise, increases computational costs, and complicates downstream trait analysis. To address this limitation, we propose a novel object-centric 3D reconstruction framework incorporating a preprocessing pipeline that leverages the Segment Anything Model v2 (SAM-2) and alpha channel background masking to achieve clean strawberry plant reconstructions. This approach produces more accurate geometric representations while substantially reducing computational time. With a background-free reconstruction, our algorithm can automatically estimate important plant traits, such as plant height and canopy width, using DBSCAN clustering and Principal Component Analysis (PCA). Experimental results show that our method outperforms conventional pipelines in both accuracy and efficiency, offering a scalable and non-destructive solution for strawberry plant phenotyping.