Cream of the Crop: Harvesting Rich, Scalable and Transferable Multi-Modal Data for Instruction Fine-Tuning

The hypothesis that pretrained large language models (LLMs) necessitate only minimal supervision during the fine-tuning (SFT) stage (Zhou et al., 2024) has been substantiated by recent advancements in data curation and selection research. However, their stability and generalizability are compromised due to the vulnerability to experimental setups and validation protocols, falling short of surpassing random sampling (Diddee & Ippolito, 2024; Xia et al., 2024b). Built upon LLMs, multi-modal LLMs (MLLMs), combined with the sheer token volume and heightened heterogeneity of data sources, amplify both the significance and complexity of data selection. To harvest multi-modal instructional data in a robust and efficient manner, we re-define the granularity of the quality metric by decomposing it into 14 vision-language-related capabilities, and introduce multi-modal rich scorers to evaluate the capabilities of each data candidate. To promote diversity, in light of the inherent objective of the alignment stage, we take interaction style as diversity indicator and use a multi-modal rich styler to identify data instruction patterns. In doing so, our multi-modal rich scorers and styler (mmSSR) guarantee that high-scoring information is conveyed to users in diversified forms. Free from embedding-based clustering or greedy sampling, mmSSR efficiently scales to millions of data with varying budget constraints, supports customization for general or specific capability acquisition, and facilitates training-free generalization to new domains for curation. Across 10+ experimental settings, validated by 14 multi-modal benchmarks, we demonstrate consistent improvements over random sampling, baseline strategies and state-of-the-art selection methods, achieving 99.1% of full performance with only 30% of the 2.6M data.

Speculative MoE: Communication Efficient Parallel MoE Inference with Speculative Token and Expert Pre-scheduling

MoE (Mixture of Experts) prevails as a neural architecture that can scale modern transformer-based LLMs (Large Language Models) to unprecedented scales. Nevertheless, large MoEs' great demands of computing power, memory capacity and memory bandwidth make scalable serving a fundamental challenge and efficient parallel inference has become a requisite to attain adequate throughput under latency constraints. DeepSpeed-MoE, one state-of-the-art MoE inference framework, adopts a 3D-parallel paradigm including EP (Expert Parallelism), TP (Tensor Parallel) and DP (Data Parallelism). However, our analysis shows DeepSpeed-MoE's inference efficiency is largely bottlenecked by EP, which is implemented with costly all-to-all collectives to route token activation. Our work aims to boost DeepSpeed-MoE by strategically reducing EP's communication overhead with a technique named Speculative MoE. Speculative MoE has two speculative parallelization schemes, speculative token shuffling and speculative expert grouping, which predict outstanding tokens' expert routing paths and pre-schedule tokens and experts across devices to losslessly trim EP's communication volume. Besides DeepSpeed-MoE, we also build Speculative MoE into a prevailing MoE inference engine SGLang. Experiments show Speculative MoE can significantly boost state-of-the-art MoE inference frameworks on fast homogeneous and slow heterogeneous interconnects.

Physics-based machine learning framework for predicting NOx emissions from compression ignition engines using on-board diagnostics data

This work presents a physics-based machine learning framework to predict and analyze oxides of nitrogen (NOx) emissions from compression-ignition engine-powered vehicles using on-board diagnostics (OBD) data as input. Accurate NOx prediction from OBD datasets is difficult because NOx formation inside an engine combustion chamber is governed by complex processes occurring on timescales much shorter than the data collection rate. Thus, emissions generally cannot be predicted accurately using simple empirically derived physics models. Black box models like genetic algorithms or neural networks can be more accurate, but have poor interpretability. The transparent model presented in this paper has both high accuracy and can explain potential sources of high emissions. The proposed framework consists of two major steps: a physics-based NOx prediction model combined with a novel Divergent Window Co-occurrence (DWC) Pattern detection algorithm to analyze operating conditions that are not adequately addressed by the physics-based model. The proposed framework is validated for generalizability with a second vehicle OBD dataset, a sensitivity analysis is performed, and model predictions are compared with that from a deep neural network. The results show that NOx emissions predictions using the proposed model has around 55% better root mean square error, and around 60% higher mean absolute error compared to the baseline NOx prediction model from previously published work. The DWC Pattern Detection Algorithm identified low engine power conditions to have high statistical significance, indicating an operating regime where the model can be improved. This work shows that the physics-based machine learning framework is a viable method for predicting NOx emissions from engines that do not incorporate NOx sensing.

Towards Universal Learning-based Model for Cardiac Image Reconstruction: Summary of the CMRxRecon2024 Challenge

Cardiovascular magnetic resonance (CMR) offers diverse imaging contrasts for assessment of cardiac function and tissue characterization. However, acquiring each single CMR modality is often time-consuming, and comprehensive clinical protocols require multiple modalities with various sampling patterns, further extending the overall acquisition time and increasing susceptibility to motion artifacts. Existing deep learning-based reconstruction methods are often designed for specific acquisition parameters, which limits their ability to generalize across a variety of scan scenarios. As part of the CMRxRecon Series, the CMRxRecon2024 challenge provides diverse datasets encompassing multi-modality multi-view imaging with various sampling patterns, and a platform for the international community to develop and benchmark reconstruction solutions in two well-crafted tasks. Task 1 is a modality-universal setting, evaluating the out-of-distribution generalization of the reconstructed model, while Task 2 follows sampling-universal setting assessing the one-for-all adaptability of the universal model. Main contributions include providing the first and largest publicly available multi-modality, multi-view cardiac k-space dataset; developing a benchmarking platform that simulates clinical acceleration protocols, with a shared code library and tutorial for various k-t undersampling patterns and data processing; giving technical insights of enhanced data consistency based on physic-informed networks and adaptive prompt-learning embedding to be versatile to different clinical settings; additional finding on evaluation metrics to address the limitations of conventional ground-truth references in universal reconstruction tasks.

Joint Transceiver Design for RIS Enhanced Dual-Functional Radar-Communication with Movable Antenna

Movable antennas (MAs) have shown significant potential in enhancing the performance of dual-functional radar-communication (DFRC) systems. In this paper, we investigate the MA-based transceiver design for DFRC systems, where a reconfigurable intelligent surface (RIS) is employed to enhance the communication quality in dead zones. To enhance the radar sensing performance, we formulate an optimization problem to maximize the radar signal-to-interference-plus-noise ratio (SINR) by jointly optimizing the beamforming vectors, receiving filter, antenna positions, and RIS reflecting coefficients. To tackle this challenging problem, we develop a fractional programming-based optimization framework, incorporating block coordinate descent (BCD), successive convex approximation (SCA), and penalty techniques. Simulation results demonstrate that the proposed method can significantly improve the radar SINR and achieve a satisfactory balance between the radar and communication performance compared with existing benchmark schemes.

A Training-Free Length Extrapolation Approach for LLMs: Greedy Attention Logit Interpolation (GALI)

Transformer-based Large Language Models (LLMs) struggle to process inputs exceeding their training context window, with performance degrading due to positional out-of-distribution (O.O.D.) that disrupt attention computations. Existing solutions, fine-tuning and training-free methods, are limited by computational inefficiency, attention logit outliers or loss of local positional information. To address this, we propose Greedy Attention Logit Interpolation (GALI), a training-free length extrapolation method that maximizes the utilization of pretrained positional intervals while avoiding attention logit outliers through attention logit interpolation. The result demonstrates that GALI consistently outperforms state-of-the-art training-free methods. Our findings reveal that LLMs interpret positional intervals unevenly within their training context window, suggesting that extrapolating within a smaller positional interval range yields superior results-even for short-context tasks. GALI represents a significant step toward resolving the positional O.O.D. challenge, enabling more reliable long-text understanding in LLMs. Our implementation of GALI, along with the experiments from our paper, is open-sourced at https://github.com/AcademyCityL/GALI.

Diffusion Model as a Noise-Aware Latent Reward Model for Step-Level Preference Optimization

Preference optimization for diffusion models aims to align them with human preferences for images. Previous methods typically leverage Vision-Language Models (VLMs) as pixel-level reward models to approximate human preferences. However, when used for step-level preference optimization, these models face challenges in handling noisy images of different timesteps and require complex transformations into pixel space. In this work, we demonstrate that diffusion models are inherently well-suited for step-level reward modeling in the latent space, as they can naturally extract features from noisy latent images. Accordingly, we propose the Latent Reward Model (LRM), which repurposes components of diffusion models to predict preferences of latent images at various timesteps. Building on LRM, we introduce Latent Preference Optimization (LPO), a method designed for step-level preference optimization directly in the latent space. Experimental results indicate that LPO not only significantly enhances performance in aligning diffusion models with general, aesthetic, and text-image alignment preferences, but also achieves 2.5-28$\times$ training speedup compared to existing preference optimization methods. Our code will be available at https://github.com/casiatao/LPO.

Co-Paced Learning Strategy Based on Confidence for Flying Bird Object Detection Model Training

To mitigate the adverse effects of hard samples on the training of the Flying Bird Object Detection (FBOD) model for surveillance videos, we propose a Co-Paced Learning Based on Confidence (CPL-BC) strategy and apply this strategy to the training process of the FBOD model. This strategy involves maintaining two models with identical structures but different initial parameter configurations, which collaborate with each other to select easy samples with prediction confidence exceeding a set threshold for training. As training progresses, the strategy gradually lowers the threshold, allowing more samples to participate, enhancing the model's ability to recognize objects from easy to hard. Before applying the CPL-BC strategy to train the FBOD models, we initially trained the two FBOD models to equip them with the capability to assess the difficulty level of flying bird object samples. Experimental results on two different datasets of flying bird objects in surveillance videos demonstrate that, compared to other model learning strategies, CPL-BC significantly improves detection accuracy, verifying the effectiveness and advancement of this method.

Fast Searching of Extreme Operating Conditions for Relay Protection Setting Calculation Based on Graph Neural Network and Reinforcement Learning

Searching for the Extreme Operating Conditions (EOCs) is one of the core problems of power system relay protection setting calculation. The current methods based on brute-force search, heuristic algorithms, and mathematical programming can hardly meet the requirements of today's power systems in terms of computation speed due to the drastic changes in operating conditions induced by renewables and power electronics. This paper proposes an EOC fast search method, named Graph Dueling Double Deep Q Network (Graph D3QN), which combines graph neural network and deep reinforcement learning to address this challenge. First, the EOC search problem is modeled as a Markov decision process, where the information of the underlying power system is extracted using graph neural networks, so that the EOC of the system can be found via deep reinforcement learning. Then, a two-stage Guided Learning and Free Exploration (GLFE) training framework is constructed to accelerate the convergence speed of reinforcement learning. Finally, the proposed Graph D3QN method is validated through case studies of searching maximum fault current for relay protection setting calculation on the IEEE 39-bus and 118-bus systems. The experimental results demonstrate that Graph D3QN can reduce the computation time by 10 to 1000 times while guaranteeing the accuracy of the selected EOCs.

A Practical Cross-Layer Approach for ML-Driven Storage Placement in Warehouse-Scale Computers

Storage systems account for a major portion of the total cost of ownership (TCO) of warehouse-scale computers, and thus have a major impact on the overall system's efficiency. Machine learning (ML)-based methods for solving key problems in storage system efficiency, such as data placement, have shown significant promise. However, there are few known practical deployments of such methods. Studying this problem in the context of real-world hyperscale data center deployments at Google, we identify a number of challenges that we believe cause this lack of practical adoption. Specifically, prior work assumes a monolithic model that resides entirely within the storage layer, an unrealistic assumption in real-world data center deployments. We propose a cross-layer approach that moves ML out of the storage system and performs it in the application running on top of it, co-designed with a scheduling algorithm at the storage layer that consumes predictions from these application-level models. This approach combines small, interpretable models with a co-designed heuristic that adapts to different online environments. We build a proof-of-concept of this approach in a production distributed computation framework at Google. Evaluations in a test deployment and large-scale simulation studies using production traces show improvements of as much as 3.47x in TCO savings compared to state of the art baselines. We believe this work represents a significant step towards more practical ML-driven storage placement in warehouse-scale computers.