Wentao Wu

Wentao Wu, Aleksei Timofeev, Chen Chen et al. · microsoft-research

We present MOFI, Manifold OF Images, a new vision foundation model designed to learn image representations from noisy entity annotated images. MOFI differs from previous work in two key aspects: (i) pre-training data, and (ii) training recipe. Regarding data, we introduce a new approach to automatically assign entity labels to images from noisy image-text pairs. Our approach involves employing a named entity recognition model to extract entities from the alt-text, and then using a CLIP model to select the correct entities as labels of the paired image. It's a simple, cost-effective method that can scale to handle billions of web-mined image-text pairs. Through this method, we have created Image-to-Entities (I2E), a new dataset with 1 billion images and 2 million distinct entities, covering rich visual concepts in the wild. Building upon the I2E dataset, we study different training recipes like supervised pre-training, contrastive pre-training, and multi-task learning. For contrastive pre-training, we treat entity names as free-form text, and further enrich them with entity descriptions. Experiments show that supervised pre-training with large-scale fine-grained entity labels is highly effective for image retrieval tasks, and multi-task training further improves the performance. The final MOFI model achieves 86.66% mAP on the challenging GPR1200 dataset, surpassing the previous state-of-the-art performance of 72.19% from OpenAI's CLIP model. Further experiments on zero-shot and linear probe image classification also show that MOFI outperforms a CLIP model trained on the original image-text data, demonstrating the effectiveness of the I2E dataset in learning strong image representations. We release our code and model weights at https://github.com/apple/ml-mofi.

Zhengfeng Lai, Haotian Zhang, Bowen Zhang et al.

Large-scale web-crawled datasets are fundamental for the success of pre-training vision-language models, such as CLIP. However, the inherent noise and potential irrelevance of web-crawled AltTexts pose challenges in achieving precise image-text alignment. Existing methods utilizing large language models (LLMs) for caption rewriting have shown promise on small, curated datasets like CC3M and CC12M. This study introduces a scalable pipeline for noisy caption rewriting. Unlike recent LLM rewriting techniques, we emphasize the incorporation of visual concepts into captions, termed as Visual-enriched Captions (VeCap). To ensure data diversity, we propose a novel mixed training scheme that optimizes the utilization of AltTexts alongside newly generated VeCap. We showcase the adaptation of this method for training CLIP on large-scale web-crawled datasets, termed VeCLIP. Employing this cost-effective pipeline, we effortlessly scale our dataset up to 300 million samples named VeCap dataset. Our results show significant advantages in image-text alignment and overall model performance. For example, VeCLIP achieves up to +25.2% gain in COCO and Flickr30k retrieval tasks under the 12M setting. For data efficiency, VeCLIP achieves +3% gain while only using 14% of the data employed in the vanilla CLIP and 11% in ALIGN. We also note the VeCap data is complementary with other well curated datasets good for zero-shot classification tasks. When combining VeCap and DFN, our model can achieve strong performance on both of image-text retrieval and zero-shot classification tasks, e.g. 83.1% accuracy@1 on ImageNet zero-shot for a H/14 model. We release the pre-trained models at https://github.com/apple/ml-veclip.

Lijie Xu, Shuang Qiu, Binhang Yuan et al. · microsoft-research

Stochastic gradient descent (SGD) is the cornerstone of modern machine learning (ML) systems. Despite its computational efficiency, SGD requires random data access that is inherently inefficient when implemented in systems that rely on block-addressable secondary storage such as HDD and SSD, e.g., TensorFlow/PyTorch and in-DB ML systems over large files. To address this impedance mismatch, various data shuffling strategies have been proposed to balance the convergence rate of SGD (which favors randomness) and its I/O performance (which favors sequential access). In this paper, we first conduct a systematic empirical study on existing data shuffling strategies, which reveals that all existing strategies have room for improvement -- they all suffer in terms of I/O performance or convergence rate. With this in mind, we propose a simple but novel hierarchical data shuffling strategy, CorgiPile. Compared with existing strategies, CorgiPile avoids a full data shuffle while maintaining comparable convergence rate of SGD as if a full shuffle were performed. We provide a non-trivial theoretical analysis of CorgiPile on its convergence behavior. We further integrate CorgiPile into PyTorch by designing new parallel/distributed shuffle operators inside a new CorgiPileDataSet API. We also integrate CorgiPile into PostgreSQL by introducing three new physical operators with optimizations. Our experimental results show that CorgiPile can achieve comparable convergence rate with the full shuffle based SGD for both deep learning and generalized linear models. For deep learning models on ImageNet dataset, CorgiPile is 1.5X faster than PyTorch with full data shuffle. For in-DB ML with linear models, CorgiPile is 1.6X-12.8X faster than two state-of-the-art in-DB ML systems, Apache MADlib and Bismarck, on both HDD and SSD.

Bojan Karlaš, David Dao, Matteo Interlandi et al. · microsoft-research

Developing modern machine learning (ML) applications is data-centric, of which one fundamental challenge is to understand the influence of data quality to ML training -- "Which training examples are 'guilty' in making the trained ML model predictions inaccurate or unfair?" Modeling data influence for ML training has attracted intensive interest over the last decade, and one popular framework is to compute the Shapley value of each training example with respect to utilities such as validation accuracy and fairness of the trained ML model. Unfortunately, despite recent intensive interest and research, existing methods only consider a single ML model "in isolation" and do not consider an end-to-end ML pipeline that consists of data transformations, feature extractors, and ML training. We present DataScope (ease.ml/datascope), the first system that efficiently computes Shapley values of training examples over an end-to-end ML pipeline, and illustrate its applications in data debugging for ML training. To this end, we first develop a novel algorithmic framework that computes Shapley value over a specific family of ML pipelines that we call canonical pipelines: a positive relational algebra query followed by a K-nearest-neighbor (KNN) classifier. We show that, for many subfamilies of canonical pipelines, computing Shapley value is in PTIME, contrasting the exponential complexity of computing Shapley value in general. We then put this to practice -- given an sklearn pipeline, we approximate it with a canonical pipeline to use as a proxy. We conduct extensive experiments illustrating different use cases and utilities. Our results show that DataScope is up to four orders of magnitude faster over state-of-the-art Monte Carlo-based methods, while being comparably, and often even more, effective in data debugging.

Tarique Siddiqui, Wentao Wu · microsoft-research

The scale and complexity of workloads in modern cloud services have brought into sharper focus a critical challenge in automated index tuning -- the need to recommend high-quality indexes while maintaining index tuning scalability. This challenge is further compounded by the requirement for automated index implementations to introduce minimal query performance regressions in production deployments, representing a significant barrier to achieving scalability and full automation. This paper directs attention to these challenges within automated index tuning and explores ways in which machine learning (ML) techniques provide new opportunities in their mitigation. In particular, we reflect on recent efforts in developing ML techniques for workload selection, candidate index filtering, speeding up index configuration search, reducing the amount of query optimizer calls, and lowering the chances of performance regressions. We highlight the key takeaways from these efforts and underline the gaps that need to be closed for their effective functioning within the traditional index tuning framework. Additionally, we present a preliminary cross-platform design aimed at democratizing index tuning across multiple SQL-like systems -- an imperative in today's continuously expanding data system landscape. We believe our findings will help provide context and impetus to the research and development efforts in automated index tuning.

Xiaoying Wang, Wentao Wu, Vivek Narasayya et al. · microsoft-research

Index tuning is critical for the performance of modern database systems. Industrial index tuners, such as the Database Tuning Advisor (DTA) developed for Microsoft SQL Server, rely on the "what-if" API provided by the query optimizer to estimate the cost of a query given an index configuration, which can lead to suboptimal recommendations when the estimations are inaccurate. Large language model (LLM) offers a new approach to index tuning, with knowledge learned from web-scale training datasets. However, the effectiveness of LLM-driven index tuning, especially beyond what is already achieved by commercial index tuners, remains unclear. In this paper, we study the practical effectiveness of LLM-driven index tuning using both industrial benchmarks and real-world enterprise customer workloads, and compare it with DTA. Our results show that although DTA is generally more reliable, with a few invocations, LLM can identify configurations that significantly outperform those found by DTA in execution time in a considerable number of cases, highlighting its potential as a complementary technique. We also observe that LLM's reasoning captures human-intuitive insights that may be distilled to potentially improve DTA. However, adopting LLM-driven index tuning in production remains challenging due to its substantial performance variance, limited and often negative impact when directly integrated into DTA, and the high cost of performance validation. This work provides motivation, lessons, and practical insights that will inspire future work on LLM-driven index tuning both in academia and industry.

Wentao Wu, Fanghua Hong, Xiao Wang et al.

Existing vehicle detectors are usually obtained by training a typical detector (e.g., YOLO, RCNN, DETR series) on vehicle images based on a pre-trained backbone (e.g., ResNet, ViT). Some researchers also exploit and enhance the detection performance using pre-trained large foundation models. However, we think these detectors may only get sub-optimal results because the large models they use are not specifically designed for vehicles. In addition, their results heavily rely on visual features, and seldom of they consider the alignment between the vehicle's semantic information and visual representations. In this work, we propose a new vehicle detection paradigm based on a pre-trained foundation vehicle model (VehicleMAE) and a large language model (T5), termed VFM-Det. It follows the region proposal-based detection framework and the features of each proposal can be enhanced using VehicleMAE. More importantly, we propose a new VAtt2Vec module that predicts the vehicle semantic attributes of these proposals and transforms them into feature vectors to enhance the vision features via contrastive learning. Extensive experiments on three vehicle detection benchmark datasets thoroughly proved the effectiveness of our vehicle detector. Specifically, our model improves the baseline approach by $+5.1\%$, $+6.2\%$ on the $AP_{0.5}$, $AP_{0.75}$ metrics, respectively, on the Cityscapes dataset.The source code of this work will be released at https://github.com/Event-AHU/VFM-Det.

Xiao Wang, Ziwen Wang, Weizhe Kong et al.

Vehicle Re-identification (Re-ID) aims to retrieve the most similar image to a given query from images captured by non-overlapping cameras. Extending vehicle Re-ID from image-only queries to text-based queries enables retrieval in real-world scenarios where only a witness description of the target vehicle is available. In this paper, we propose PFCVR, a Part-level Fine-grained Cross-modal Vehicle Retrieval model for text-to-image vehicle re-identification. PFCVR constructs locally paired images and texts at the part level and introduces learnable part-query tokens that aggregate both part-specific and full-sentence context before aligning with visual part features. On top of this explicit local alignment, a bi-directional mask recovery module lets each modality reconstruct its masked content under the guidance of the other, implicitly bridging local correspondences into global feature alignment. Furthermore, we construct a new large-scale dataset called T2I-VeRW, which contains 14,668 images covering 1,796 vehicle identities with fine-grained part-level annotations. Experimental results on the T2I-VeRI dataset show that PFCVR achieves 29.2\% Rank-1 accuracy, improving over the best competing method by +3.7\% percentage points. On the newly proposed T2I-VeRW benchmark, PFCVR achieves 55.2\% Rank-1 accuracy, outperforming a comprehensive set of recent state-of-the-art methods. Source code will be released on https://github.com/Event-AHU/Neuromorphic_ReID

Xiao Wang, Shiao Wang, Yuhe Ding et al.

In the post-deep learning era, the Transformer architecture has demonstrated its powerful performance across pre-trained big models and various downstream tasks. However, the enormous computational demands of this architecture have deterred many researchers. To further reduce the complexity of attention models, numerous efforts have been made to design more efficient methods. Among them, the State Space Model (SSM), as a possible replacement for the self-attention based Transformer model, has drawn more and more attention in recent years. In this paper, we give the first comprehensive review of these works and also provide experimental comparisons and analysis to better demonstrate the features and advantages of SSM. Specifically, we first give a detailed description of principles to help the readers quickly capture the key ideas of SSM. After that, we dive into the reviews of existing SSMs and their various applications, including natural language processing, computer vision, graph, multi-modal and multi-media, point cloud/event stream, time series data, and other domains. In addition, we give statistical comparisons and analysis of these models and hope it helps the readers to understand the effectiveness of different structures on various tasks. Then, we propose possible research points in this direction to better promote the development of the theoretical model and application of SSM. More related works will be continuously updated on the following GitHub: https://github.com/Event-AHU/Mamba_State_Space_Model_Paper_List.

Tom Gunter, Zirui Wang, Chong Wang et al.

We present foundation language models developed to power Apple Intelligence features, including a ~3 billion parameter model designed to run efficiently on devices and a large server-based language model designed for Private Cloud Compute. These models are designed to perform a wide range of tasks efficiently, accurately, and responsibly. This report describes the model architecture, the data used to train the model, the training process, how the models are optimized for inference, and the evaluation results. We highlight our focus on Responsible AI and how the principles are applied throughout the model development.

Xiao Wang, Yu Jin, Wentao Wu et al.

Object detection in event streams has emerged as a cutting-edge research area, demonstrating superior performance in low-light conditions, scenarios with motion blur, and rapid movements. Current detectors leverage spiking neural networks, Transformers, or convolutional neural networks as their core architectures, each with its own set of limitations including restricted performance, high computational overhead, or limited local receptive fields. This paper introduces a novel MoE (Mixture of Experts) heat conduction-based object detection algorithm that strikingly balances accuracy and computational efficiency. Initially, we employ a stem network for event data embedding, followed by processing through our innovative MoE-HCO blocks. Each block integrates various expert modules to mimic heat conduction within event streams. Subsequently, an IoU-based query selection module is utilized for efficient token extraction, which is then channeled into a detection head for the final object detection process. Furthermore, we are pleased to introduce EvDET200K, a novel benchmark dataset for event-based object detection. Captured with a high-definition Prophesee EVK4-HD event camera, this dataset encompasses 10 distinct categories, 200,000 bounding boxes, and 10,054 samples, each spanning 2 to 5 seconds. We also provide comprehensive results from over 15 state-of-the-art detectors, offering a solid foundation for future research and comparison. The source code of this paper will be released on: https://github.com/Event-AHU/OpenEvDET

Xiao Wang, Wentao Wu, Chenglong Li et al.

Understanding vehicles in images is important for various applications such as intelligent transportation and self-driving system. Existing vehicle-centric works typically pre-train models on large-scale classification datasets and then fine-tune them for specific downstream tasks. However, they neglect the specific characteristics of vehicle perception in different tasks and might thus lead to sub-optimal performance. To address this issue, we propose a novel vehicle-centric pre-training framework called VehicleMAE, which incorporates the structural information including the spatial structure from vehicle profile information and the semantic structure from informative high-level natural language descriptions for effective masked vehicle appearance reconstruction. To be specific, we explicitly extract the sketch lines of vehicles as a form of the spatial structure to guide vehicle reconstruction. The more comprehensive knowledge distilled from the CLIP big model based on the similarity between the paired/unpaired vehicle image-text sample is further taken into consideration to help achieve a better understanding of vehicles. A large-scale dataset is built to pre-train our model, termed Autobot1M, which contains about 1M vehicle images and 12693 text information. Extensive experiments on four vehicle-based downstream tasks fully validated the effectiveness of our VehicleMAE. The source code and pre-trained models will be released at https://github.com/Event-AHU/VehicleMAE.

Xiao Wang, Yuehang Li, Wentao Wu et al.

Existing X-ray based pre-trained vision models are usually conducted on a relatively small-scale dataset (less than 500k samples) with limited resolution (e.g., 224 $\times$ 224). However, the key to the success of self-supervised pre-training large models lies in massive training data, and maintaining high resolution in the field of X-ray images is the guarantee of effective solutions to difficult miscellaneous diseases. In this paper, we address these issues by proposing the first high-definition (1280 $\times$ 1280) X-ray based pre-trained foundation vision model on our newly collected large-scale dataset which contains more than 1 million X-ray images. Our model follows the masked auto-encoder framework which takes the tokens after mask processing (with a high rate) is used as input, and the masked image patches are reconstructed by the Transformer encoder-decoder network. More importantly, we introduce a novel context-aware masking strategy that utilizes the chest contour as a boundary for adaptive masking operations. We validate the effectiveness of our model on two downstream tasks, including X-ray report generation and disease recognition. Extensive experiments demonstrate that our pre-trained medical foundation vision model achieves comparable or even new state-of-the-art performance on downstream benchmark datasets. The source code and pre-trained models of this paper will be released on https://github.com/Event-AHU/Medical_Image_Analysis.

Wentao Wu, Xiao Wang, Chenglong Li et al.

Vehicle-centric perception plays a crucial role in many intelligent systems, including large-scale surveillance systems, intelligent transportation, and autonomous driving. Existing approaches lack effective learning of vehicle-related knowledge during pre-training, resulting in poor capability for modeling general vehicle perception representations. To handle this problem, we propose VehicleMAE-V2, a novel vehicle-centric pre-trained large model. By exploring and exploiting vehicle-related multimodal structured priors to guide the masked token reconstruction process, our approach can significantly enhance the model's capability to learn generalizable representations for vehicle-centric perception. Specifically, we design the Symmetry-guided Mask Module (SMM), Contour-guided Representation Module (CRM) and Semantics-guided Representation Module (SRM) to incorporate three kinds of structured priors into token reconstruction including symmetry, contour and semantics of vehicles respectively. SMM utilizes the vehicle symmetry constraints to avoid retaining symmetric patches and can thus select high-quality masked image patches and reduce information redundancy. CRM minimizes the probability distribution divergence between contour features and reconstructed features and can thus preserve holistic vehicle structure information during pixel-level reconstruction. SRM aligns image-text features through contrastive learning and cross-modal distillation to address the feature confusion caused by insufficient semantic understanding during masked reconstruction. To support the pre-training of VehicleMAE-V2, we construct Autobot4M, a large-scale dataset comprising approximately 4 million vehicle images and 12,693 text descriptions. Extensive experiments on five downstream tasks demonstrate the superior performance of VehicleMAE-V2.

Long Phan, Alice Gatti, Ziwen Han et al. · amazon-science, apple-ml

Benchmarks are important tools for tracking the rapid advancements in large language model (LLM) capabilities. However, benchmarks are not keeping pace in difficulty: LLMs now achieve over 90\% accuracy on popular benchmarks like MMLU, limiting informed measurement of state-of-the-art LLM capabilities. In response, we introduce Humanity's Last Exam (HLE), a multi-modal benchmark at the frontier of human knowledge, designed to be the final closed-ended academic benchmark of its kind with broad subject coverage. HLE consists of 2,500 questions across dozens of subjects, including mathematics, humanities, and the natural sciences. HLE is developed globally by subject-matter experts and consists of multiple-choice and short-answer questions suitable for automated grading. Each question has a known solution that is unambiguous and easily verifiable, but cannot be quickly answered via internet retrieval. State-of-the-art LLMs demonstrate low accuracy and calibration on HLE, highlighting a significant gap between current LLM capabilities and the expert human frontier on closed-ended academic questions. To inform research and policymaking upon a clear understanding of model capabilities, we publicly release HLE at https://lastexam.ai.

Wentao Wu, Xiao Wang, Chenglong Li et al.

Event cameras have attracted increasing attention in recent years due to their advantages in high dynamic range, high temporal resolution, low power consumption, and low latency. Some researchers have begun exploring pre-training directly on event data. Nevertheless, these efforts often fail to establish strong connections with RGB frames, limiting their applicability in multi-modal fusion scenarios. To address these issues, we propose a novel CM3AE pre-training framework for the RGB-Event perception. This framework accepts multi-modalities/views of data as input, including RGB images, event images, and event voxels, providing robust support for both event-based and RGB-event fusion based downstream tasks. Specifically, we design a multi-modal fusion reconstruction module that reconstructs the original image from fused multi-modal features, explicitly enhancing the model's ability to aggregate cross-modal complementary information. Additionally, we employ a multi-modal contrastive learning strategy to align cross-modal feature representations in a shared latent space, which effectively enhances the model's capability for multi-modal understanding and capturing global dependencies. We construct a large-scale dataset containing 2,535,759 RGB-Event data pairs for the pre-training. Extensive experiments on five downstream tasks fully demonstrated the effectiveness of CM3AE. Source code and pre-trained models will be released on https://github.com/Event-AHU/CM3AE.

Xiao Wang, Qingquan Yang, Fuling Wang et al.

Nuclear fusion is one of the most promising ways for humans to obtain infinite energy. Currently, with the rapid development of artificial intelligence, the mission of nuclear fusion has also entered a critical period of its development. How to let more people to understand nuclear fusion and join in its research is one of the effective means to accelerate the implementation of fusion. This paper proposes the first large model in the field of nuclear fusion, XiHeFusion, which is obtained through supervised fine-tuning based on the open-source large model Qwen2.5-14B. We have collected multi-source knowledge about nuclear fusion tasks to support the training of this model, including the common crawl, eBooks, arXiv, dissertation, etc. After the model has mastered the knowledge of the nuclear fusion field, we further used the chain of thought to enhance its logical reasoning ability, making XiHeFusion able to provide more accurate and logical answers. In addition, we propose a test questionnaire containing 180+ questions to assess the conversational ability of this science popularization large model. Extensive experimental results show that our nuclear fusion dialogue model, XiHeFusion, can perform well in answering science popularization knowledge. The pre-trained XiHeFusion model is released on https://github.com/Event-AHU/XiHeFusion.

Xiao Wang, Ziwen Wang, Wentao Wu et al.

With the rapid advancement of autonomous driving, vehicle perception, particularly detection and segmentation, has placed increasingly higher demands on algorithmic performance. Pre-trained large segmentation models, especially Segment Anything Model (SAM), have sparked significant interest and inspired new research directions in artificial intelligence. However, SAM cannot be directly applied to the fine-grained task of vehicle part segmentation, as its text-prompted segmentation functionality is not publicly accessible, and the mask regions generated by its default mode lack semantic labels, limiting its utility in structured, category-specific segmentation tasks. To address these limitations, we propose SAV, a novel framework comprising three core components: a SAM-based encoder-decoder, a vehicle part knowledge graph, and a context sample retrieval encoding module. The knowledge graph explicitly models the spatial and geometric relationships among vehicle parts through a structured ontology, effectively encoding prior structural knowledge. Meanwhile, the context retrieval module enhances segmentation by identifying and leveraging visually similar vehicle instances from training data, providing rich contextual priors for improved generalization. Furthermore, we introduce a new large-scale benchmark dataset for vehicle part segmentation, named VehicleSeg10K, which contains 11,665 high-quality pixel-level annotations across diverse scenes and viewpoints. We conduct comprehensive experiments on this dataset and two other datasets, benchmarking multiple representative baselines to establish a solid foundation for future research and comparison. % Both the dataset and source code of this paper will be released upon acceptance. Both the dataset and source code of this paper will be released on https://github.com/Event-AHU/SAV

Hanmeng Zhong, Linqing Chen, Wentao Wu et al.

Recent development in Retrieval-Augmented Large Language Models (LLMs) have shown great promise in biomedical applications. How ever, a critical gap persists in reliably evaluating their curation ability the process by which models select and integrate relevant references while filtering out noise. To address this, we introduce the benchmark for Curation of Retrieval-Augmented LLMs in Biomedicine (CRAB), the first multilingual benchmark tailored for evaluating the biomedical curation of retrieval-augmented LLMs, available in English, French, German and Chinese. By incorporating a novel citation-based evaluation metric, CRAB quantifies the curation performance of retrieval-augmented LLMs in biomedicine. Experimental results reveal significant discrepancies in the curation performance of mainstream LLMs, underscoring the urgent need to improve it in the domain of biomedicine. Our dataset is available at https://huggingface.co/datasets/zhm0/CRAB.

Yang Li, Yu Shen, Wentao Zhang et al.

Black-box optimization (BBO) has a broad range of applications, including automatic machine learning, engineering, physics, and experimental design. However, it remains a challenge for users to apply BBO methods to their problems at hand with existing software packages, in terms of applicability, performance, and efficiency. In this paper, we build OpenBox, an open-source and general-purpose BBO service with improved usability. The modular design behind OpenBox also facilitates flexible abstraction and optimization of basic BBO components that are common in other existing systems. OpenBox is distributed, fault-tolerant, and scalable. To improve efficiency, OpenBox further utilizes "algorithm agnostic" parallelization and transfer learning. Our experimental results demonstrate the effectiveness and efficiency of OpenBox compared to existing systems.

Yu Liu, Hantian Zhang, Luyuan Zeng et al.

We conduct an empirical study of machine learning functionalities provided by major cloud service providers, which we call machine learning clouds. Machine learning clouds hold the promise of hiding all the sophistication of running large-scale machine learning: Instead of specifying how to run a machine learning task, users only specify what machine learning task to run and the cloud figures out the rest. Raising the level of abstraction, however, rarely comes free - a performance penalty is possible. How good, then, are current machine learning clouds on real-world machine learning workloads? We study this question with a focus on binary classication problems. We present mlbench, a novel benchmark constructed by harvesting datasets from Kaggle competitions. We then compare the performance of the top winning code available from Kaggle with that of running machine learning clouds from both Azure and Amazon on mlbench. Our comparative study reveals the strength and weakness of existing machine learning clouds and points out potential future directions for improvement.

Ethan Li, Anders Boesen Lindbo Larsen, Chen Zhang et al. · apple-ml, cmu

We introduce two multilingual, multimodal foundation language models that power Apple Intelligence features across Apple devices and services: i a 3B-parameter on-device model optimized for Apple silicon through architectural innovations such as KV-cache sharing and 2-bit quantization-aware training; and ii a scalable server model built on a novel Parallel-Track Mixture-of-Experts PT-MoE transformer that combines track parallelism, mixture-of-experts sparse computation, and interleaved global-local attention to deliver high quality with competitive cost on Apple's Private Cloud Compute platform. Both models are trained on large-scale multilingual and multimodal datasets sourced via responsible web crawling, licensed corpora, and high-quality synthetic data, then further refined with supervised fine-tuning and reinforcement learning on a new asynchronous platform. The resulting models support several additional languages while understanding images and executing tool calls. In public benchmarks and human evaluations, both the server model and the on-device model match or surpass comparably sized open baselines. A new Swift-centric Foundation Models framework exposes guided generation, constrained tool calling, and LoRA adapter fine-tuning, allowing developers to integrate these capabilities with a few lines of code. The latest advancements in Apple Intelligence models are grounded in our Responsible AI approach with safeguards like content filtering and locale-specific evaluation, as well as our commitment to protecting our users' privacy with innovations like Private Cloud Compute.

Wentao Wu, Chi Wang · microsoft-research

Modern database systems rely on cost-based query optimizers to come up with good execution plans for input queries. Such query optimizers rely on cost models to estimate the costs of candidate query execution plans. A cost model represents a function from a set of cost units to query execution cost, where each cost unit specifies the unit cost of executing a certain type of query processing operation (such as table scan or join). These cost units are traditionally viewed as constants, whose values only depend on the platform configuration where the database system runs on top of but are invariant for queries processed by the database system. In this paper, we challenge this classic view by thinking of these cost units as variables instead. We show that, by varying the cost-unit values one can obtain query plans that significantly outperform the default query plans returned by the query optimizer when viewing the cost units as constants. We term this cost-unit tuning process "query tuning" (QT) and show that it is similar to the well-known hyper-parameter optimization (HPO) problem in AutoML. As a result, any state-of-the-art HPO technologies can be applied to QT. We study the QT problem in the context of anytime tuning, which is desirable in practice by constraining the total time spent on QT within a given budget -- we call this problem budget-aware query tuning. We further extend our study from tuning a single query to tuning a workload with multiple queries, and we call this generalized problem budget-aware workload tuning (WT), which aims for minimizing the execution time of the entire workload. WT is more challenging as one needs to further prioritize individual query tuning within the given time budget. We propose solutions to both QT and WT and experimental evaluation using both benchmark and real workloads demonstrates the efficacy of our proposed solutions.

Linqing Chen, Weilei Wang, Yubin Xia et al.

In this paper, we propose a novel system that integrates state-of-the-art, domain-specific large language models with advanced information retrieval techniques to deliver comprehensive and context-aware responses. Our approach facilitates seamless interaction among diverse components, enabling cross-validation of outputs to produce accurate, high-quality responses enriched with relevant data, images, tables, and other modalities. We demonstrate the system's capability to enhance response precision by leveraging a robust question-answering model, significantly improving the quality of dialogue generation. The system provides an accessible platform for real-time, high-fidelity interactions, allowing users to benefit from efficient human-computer interaction, precise retrieval, and simultaneous access to a wide range of literature and data. This dramatically improves the research efficiency of professionals in the biomedical and pharmaceutical domains and facilitates faster, more informed decision-making throughout the R\&D process. Furthermore, the system proposed in this paper is available at https://synapse-chat.patsnap.com.

Linqing Chen, Hanmeng Zhong, Wentao Wu et al.

Large language model (LLM) training faces a critical bottleneck: the scarcity of high-quality, reasoning-intensive question-answer pairs, especially from sparse, domain-specific sources like PubMed papers or legal documents. Existing methods rely on surface patterns, fundamentally failing to generate controllable, complex multi-hop reasoning questions that test genuine understanding-essential for advancing LLM training paradigms. We present \textbf{Semantic Bridge}, the first universal framework for controllably generating sophisticated multi-hop reasoning questions from arbitrary sources. Our breakthrough innovation is \textit{semantic graph weaving}-three complementary bridging mechanisms (entity bridging for role-varying shared entities, predicate chain bridging for temporal/causal/logical sequences, and causal bridging for explicit reasoning chains)-that systematically construct complex pathways across documents, with fine-grained control over complexity and types via AMR-driven analysis. Our multi-modal AMR pipeline achieves up to 9.5% better round-trip quality, enabling production-ready controllable QA generation. Extensive evaluation demonstrates performance across both general-purpose datasets (Wikipedia) and specialized domains (biomedicine) It yields consistent 18.3%-25.4% gains over baselines across four languages (English, Chinese, French, German). Question pairs generated from 200 sources outperform 600 native human annotation examples with 67% fewer materials. Human evaluation shows 23.4% higher complexity, 18.7% better answerability, and 31.2% improved pattern coverage. Semantic Bridge establishes a new paradigm for LLM training data synthesis, enabling controllable generation of targeted reasoning questions from sparse sources. We will release our core code and semantic bridge model.

Wentao Wu, Chenglong Li, Xiao Wang et al.

Existing multimodal UAV object detection methods often overlook the impact of semantic gaps between modalities, which makes it difficult to achieve accurate semantic and spatial alignments, limiting detection performance. To address this problem, we propose a Large Language Model (LLM) guided Progressive feature Alignment Network called LPANet, which leverages the semantic features extracted from a large language model to guide the progressive semantic and spatial alignment between modalities for multimodal UAV object detection. To employ the powerful semantic representation of LLM, we generate the fine-grained text descriptions of each object category by ChatGPT and then extract the semantic features using the large language model MPNet. Based on the semantic features, we guide the semantic and spatial alignments in a progressive manner as follows. First, we design the Semantic Alignment Module (SAM) to pull the semantic features and multimodal visual features of each object closer, alleviating the semantic differences of objects between modalities. Second, we design the Explicit Spatial alignment Module (ESM) by integrating the semantic relations into the estimation of feature-level offsets, alleviating the coarse spatial misalignment between modalities. Finally, we design the Implicit Spatial alignment Module (ISM), which leverages the cross-modal correlations to aggregate key features from neighboring regions to achieve implicit spatial alignment. Comprehensive experiments on two public multimodal UAV object detection datasets demonstrate that our approach outperforms state-of-the-art multimodal UAV object detectors.

Wentao Wu, Linqing Chen, Hanmeng Zhong et al.

BalancedBio is a theoretically grounded framework for parameter-efficient biomedical reasoning, addressing multi-capability integration in domain-specific AI alignment. It establishes the Biomedical Multi-Capability Convergence Theorem, proving orthogonal gradient spaces are essential to prevent capability interference for safe deployment. Key innovations include: (1) Medical Knowledge Grounded Synthetic Generation (MKGSG), extending Source2Synth with clinical workflow constraints and medical ontology validation for factual accuracy and safety; and (2) Capability Aware Group Relative Policy Optimization, deriving optimal hybrid reward weighting to maintain orthogonality in RL, using a reward model with rule-based and model-based scores adapted to biomedical tasks. Mathematical analysis proves Pareto-optimal convergence, preserving performance across capabilities. It achieves state-of-the-art results in its parameter class: domain expertise (80.95% BIOMED-MMLU, +15.32% over baseline), reasoning (61.94%, +7.75%), instruction following (67.95%, +6.44%), and integration (86.7%, +18.5%). Theoretical safety guarantees include bounds on capability preservation and clinical accuracy. Real-world deployment yields 78% cost reduction, 23% improved diagnostic accuracy, and 89% clinician acceptance. This work provides a principled methodology for biomedical AI alignment, enabling efficient reasoning with essential safety and reliability, with the 0.5B model version to be released.

Linqing Chen, Weilei Wang, Wentao Wu et al.

Hierarchical Extreme Multi-Label Classification poses greater difficulties compared to traditional multi-label classification because of the intricate hierarchical connections of labels within a domain-specific taxonomy and the substantial number of labels. Some of the prior research endeavors centered on classifying text through several ancillary stages such as the cluster algorithm and multiphase classification. Others made attempts to leverage the assistance of generative methods yet were unable to properly control the output of the generative model. We redefine the task from hierarchical multi-Label classification to Hierarchical Multi-Label Generation (HMG) and employ a generative framework with Probabilistic Level Constraints (PLC) to generate hierarchical labels within a specific taxonomy that have complex hierarchical relationships. The approach we proposed in this paper enables the framework to generate all relevant labels across levels for each document without relying on preliminary operations like clustering. Meanwhile, it can control the model output precisely in terms of count, length, and level aspects. Experiments demonstrate that our approach not only achieves a new SOTA performance in the HMG task, but also has a much better performance in constrained the output of model than previous research work.

Xinnan Zhu, Yicheng Zhu, Tixin Chen et al.

Temporal action detection aims to locate and classify actions in untrimmed videos. While recent works focus on designing powerful feature processors for pre-trained representations, they often overlook the inherent noise and redundancy within these features. Large-scale pre-trained video encoders tend to introduce background clutter and irrelevant semantics, leading to context confusion and imprecise boundaries. To address this, we propose a frequency-aware decoupling network that improves action discriminability by filtering out noisy semantics captured by pre-trained models. Specifically, we introduce an adaptive temporal decoupling scheme that suppresses irrelevant information while preserving fine-grained atomic action details, yielding more task-specific representations. In addition, we enhance inter-frame modeling by capturing temporal variations to better distinguish actions from background redundancy. Furthermore, we present a long-short-term category-aware relation network that jointly models local transitions and long-range dependencies, improving localization precision. The refined atomic features and frequency-guided dynamics are fed into a standard detection head to produce accurate action predictions. Extensive experiments on THUMOS14, HACS, and ActivityNet-1.3 show that our method, powered by InternVideo2-6B features, achieves state-of-the-art performance on temporal action detection benchmarks.

Linqing Chen, Weilei Wang, Zilong Bai et al.

Large language models (LLMs) have revolutionized Natural Language Processing (NLP) by minimizing the need for complex feature engineering. However, the application of LLMs in specialized domains like biopharmaceuticals and chemistry remains largely unexplored. These fields are characterized by intricate terminologies, specialized knowledge, and a high demand for precision areas where general purpose LLMs often fall short. In this study, we introduce PharmaGPT, a suite of domain specilized LLMs with 13 billion and 70 billion parameters, specifically trained on a comprehensive corpus tailored to the Bio-Pharmaceutical and Chemical domains. Our evaluation shows that PharmaGPT surpasses existing general models on specific-domain benchmarks such as NAPLEX, demonstrating its exceptional capability in domain-specific tasks. Remarkably, this performance is achieved with a model that has only a fraction, sometimes just one-tenth-of the parameters of general-purpose large models. This advancement establishes a new benchmark for LLMs in the bio-pharmaceutical and chemical fields, addressing the existing gap in specialized language modeling. It also suggests a promising path for enhanced research and development, paving the way for more precise and effective NLP applications in these areas.

Lijie Xu, Chulin Xie, Yiran Guo et al. · microsoft-research

Current federated learning (FL) approaches view decentralized training data as a single table, divided among participants either horizontally (by rows) or vertically (by columns). However, these approaches are inadequate for handling distributed relational tables across databases. This scenario requires intricate SQL operations like joins and unions to obtain the training data, which is either costly or restricted by privacy concerns. This raises the question: can we directly run FL on distributed relational tables? In this paper, we formalize this problem as relational federated learning (RFL). We propose TablePuppet, a generic framework for RFL that decomposes the learning process into two steps: (1) learning over join (LoJ) followed by (2) learning over union (LoU). In a nutshell, LoJ pushes learning down onto the vertical tables being joined, and LoU further pushes learning down onto the horizontal partitions of each vertical table. TablePuppet incorporates computation/communication optimizations to deal with the duplicate tuples introduced by joins, as well as differential privacy (DP) to protect against both feature and label leakages. We demonstrate the efficiency of TablePuppet in combination with two widely-used ML training algorithms, stochastic gradient descent (SGD) and alternating direction method of multipliers (ADMM), and compare their computation/communication complexity. We evaluate the SGD/ADMM algorithms developed atop TablePuppet by training diverse ML models. Our experimental results show that TablePuppet achieves model accuracy comparable to the centralized baselines running directly atop the SQL results. Moreover, ADMM takes less communication time than SGD to converge to similar model accuracy.

Yang Li, Yu Shen, Wentao Zhang et al.

End-to-end AutoML has attracted intensive interests from both academia and industry, which automatically searches for ML pipelines in a space induced by feature engineering, algorithm/model selection, and hyper-parameter tuning. Existing AutoML systems, however, suffer from scalability issues when applying to application domains with large, high-dimensional search spaces. We present VolcanoML, a scalable and extensible framework that facilitates systematic exploration of large AutoML search spaces. VolcanoML introduces and implements basic building blocks that decompose a large search space into smaller ones, and allows users to utilize these building blocks to compose an execution plan for the AutoML problem at hand. VolcanoML further supports a Volcano-style execution model - akin to the one supported by modern database systems - to execute the plan constructed. Our evaluation demonstrates that, not only does VolcanoML raise the level of expressiveness for search space decomposition in AutoML, it also leads to actual findings of decomposition strategies that are significantly more efficient than the ones employed by state-of-the-art AutoML systems such as auto-sklearn.

Jiawei Jiang, Shaoduo Gan, Yue Liu et al.

The appeal of serverless (FaaS) has triggered a growing interest on how to use it in data-intensive applications such as ETL, query processing, or machine learning (ML). Several systems exist for training large-scale ML models on top of serverless infrastructures (e.g., AWS Lambda) but with inconclusive results in terms of their performance and relative advantage over "serverful" infrastructures (IaaS). In this paper we present a systematic, comparative study of distributed ML training over FaaS and IaaS. We present a design space covering design choices such as optimization algorithms and synchronization protocols, and implement a platform, LambdaML, that enables a fair comparison between FaaS and IaaS. We present experimental results using LambdaML, and further develop an analytic model to capture cost/performance tradeoffs that must be considered when opting for a serverless infrastructure. Our results indicate that ML training pays off in serverless only for models with efficient (i.e., reduced) communication and that quickly converge. In general, FaaS can be much faster but it is never significantly cheaper than IaaS.

Cedric Renggli, Luka Rimanic, Nezihe Merve Gürel et al.

Developing machine learning models can be seen as a process similar to the one established for traditional software development. A key difference between the two lies in the strong dependency between the quality of a machine learning model and the quality of the data used to train or perform evaluations. In this work, we demonstrate how different aspects of data quality propagate through various stages of machine learning development. By performing a joint analysis of the impact of well-known data quality dimensions and the downstream machine learning process, we show that different components of a typical MLOps pipeline can be efficiently designed, providing both a technical and theoretical perspective.

Cedric Renggli, Luka Rimanic, Luka Kolar et al.

In our experience of working with domain experts who are using today's AutoML systems, a common problem we encountered is what we call "unrealistic expectations" -- when users are facing a very challenging task with a noisy data acquisition process, while being expected to achieve startlingly high accuracy with machine learning (ML). Many of these are predestined to fail from the beginning. In traditional software engineering, this problem is addressed via a feasibility study, an indispensable step before developing any software system. In this paper, we present Snoopy, with the goal of supporting data scientists and machine learning engineers performing a systematic and theoretically founded feasibility study before building ML applications. We approach this problem by estimating the irreducible error of the underlying task, also known as the Bayes error rate (BER), which stems from data quality issues in datasets used to train or evaluate ML model artifacts. We design a practical Bayes error estimator that is compared against baseline feasibility study candidates on 6 datasets (with additional real and synthetic noise of different levels) in computer vision and natural language processing. Furthermore, by including our systematic feasibility study with additional signals into the iterative label cleaning process, we demonstrate in end-to-end experiments how users are able to save substantial labeling time and monetary efforts.

Bojan Karlaš, Peng Li, Renzhi Wu et al.

Machine learning (ML) applications have been thriving recently, largely attributed to the increasing availability of data. However, inconsistency and incomplete information are ubiquitous in real-world datasets, and their impact on ML applications remains elusive. In this paper, we present a formal study of this impact by extending the notion of Certain Answers for Codd tables, which has been explored by the database research community for decades, into the field of machine learning. Specifically, we focus on classification problems and propose the notion of "Certain Predictions" (CP) -- a test data example can be certainly predicted (CP'ed) if all possible classifiers trained on top of all possible worlds induced by the incompleteness of data would yield the same prediction. We study two fundamental CP queries: (Q1) checking query that determines whether a data example can be CP'ed; and (Q2) counting query that computes the number of classifiers that support a particular prediction (i.e., label). Given that general solutions to CP queries are, not surprisingly, hard without assumption over the type of classifier, we further present a case study in the context of nearest neighbor (NN) classifiers, where efficient solutions to CP queries can be developed -- we show that it is possible to answer both queries in linear or polynomial time over exponentially many possible worlds. We demonstrate one example use case of CP in the important application of "data cleaning for machine learning (DC for ML)." We show that our proposed CPClean approach built based on CP can often significantly outperform existing techniques in terms of classification accuracy with mild manual cleaning effort.

Fotis Psallidas, Yiwen Zhu, Bojan Karlas et al.

The recent success of machine learning (ML) has led to an explosive growth both in terms of new systems and algorithms built in industry and academia, and new applications built by an ever-growing community of data science (DS) practitioners. This quickly shifting panorama of technologies and applications is challenging for builders and practitioners alike to follow. In this paper, we set out to capture this panorama through a wide-angle lens, by performing the largest analysis of DS projects to date, focusing on questions that can help determine investments on either side. Specifically, we download and analyze: (a) over 6M Python notebooks publicly available on GITHUB, (b) over 2M enterprise DS pipelines developed within COMPANYX, and (c) the source code and metadata of over 900 releases from 12 important DS libraries. The analysis we perform ranges from coarse-grained statistical characterizations to analysis of library imports, pipelines, and comparative studies across datasets and time. We report a large number of measurements for our readers to interpret, and dare to draw a few (actionable, yet subjective) conclusions on (a) what systems builders should focus on to better serve practitioners, and (b) what technologies should practitioners bet on given current trends. We plan to automate this analysis and release associated tools and results periodically.

Frances Ann Hubis, Wentao Wu, Ce Zhang

Simplifying machine learning (ML) application development, including distributed computation, programming interface, resource management, model selection, etc, has attracted intensive interests recently. These research efforts have significantly improved the efficiency and the degree of automation of developing ML models. In this paper, we take a first step in an orthogonal direction towards automated quality management for human-in-the-loop ML application development. We build ease. ml/meter, a system that can automatically detect and measure the degree of overfitting during the whole lifecycle of ML application development. ease. ml/meter returns overfitting signals with strong probabilistic guarantees, based on which developers can take appropriate actions. In particular, ease. ml/meter provides principled guidelines to simple yet nontrivial questions regarding desired validation and test data sizes, which are among commonest questions raised by developers. The fact that ML application development is typically a continuous procedure further worsens the situation: The validation and test data sets can lose their statistical power quickly due to multiple accesses, especially in the presence of adaptive analysis. ease. ml/meter addresses these challenges by leveraging a collection of novel techniques and optimizations, resulting in practically tractable data sizes without compromising the probabilistic guarantees. We present the design and implementation details of ease. ml/meter, as well as detailed theoretical analysis and empirical evaluation of its effectiveness.

Cedric Renggli, Bojan Karlaš, Bolin Ding et al.

Continuous integration is an indispensable step of modern software engineering practices to systematically manage the life cycles of system development. Developing a machine learning model is no difference - it is an engineering process with a life cycle, including design, implementation, tuning, testing, and deployment. However, most, if not all, existing continuous integration engines do not support machine learning as first-class citizens. In this paper, we present ease.ml/ci, to our best knowledge, the first continuous integration system for machine learning. The challenge of building ease.ml/ci is to provide rigorous guarantees, e.g., single accuracy point error tolerance with 0.999 reliability, with a practical amount of labeling effort, e.g., 2K labels per test. We design a domain specific language that allows users to specify integration conditions with reliability constraints, and develop simple novel optimizations that can lower the number of labels required by up to two orders of magnitude for test conditions popularly used in real production systems.

Tian Li, Jie Zhong, Ji Liu et al.

We present ease.ml, a declarative machine learning service platform we built to support more than ten research groups outside the computer science departments at ETH Zurich for their machine learning needs. With ease.ml, a user defines the high-level schema of a machine learning application and submits the task via a Web interface. The system automatically deals with the rest, such as model selection and data movement. In this paper, we describe the ease.ml architecture and focus on a novel technical problem introduced by ease.ml regarding resource allocation. We ask, as a "service provider" that manages a shared cluster of machines among all our users running machine learning workloads, what is the resource allocation strategy that maximizes the global satisfaction of all our users? Resource allocation is a critical yet subtle issue in this multi-tenant scenario, as we have to balance between efficiency and fairness. We first formalize the problem that we call multi-tenant model selection, aiming for minimizing the total regret of all users running automatic model selection tasks. We then develop a novel algorithm that combines multi-armed bandits with Bayesian optimization and prove a regret bound under the multi-tenant setting. Finally, we report our evaluation of ease.ml on synthetic data and on one service we are providing to our users, namely, image classification with deep neural networks. Our experimental evaluation results show that our proposed solution can be up to 9.8x faster in achieving the same global quality for all users as the two popular heuristics used by our users before ease.ml.

Xi Wu, Matthew Fredrikson, Wentao Wu et al.

Private regression has received attention from both database and security communities. Recent work by Fredrikson et al. (USENIX Security 2014) analyzed the functional mechanism (Zhang et al. VLDB 2012) for training linear regression models over medical data. Unfortunately, they found that model accuracy is already unacceptable with differential privacy when $\varepsilon = 5$. We address this issue, presenting an explicit connection between differential privacy and stable learning theory through which a substantially better privacy/utility tradeoff can be obtained. Perhaps more importantly, our theory reveals that the most basic mechanism in differential privacy, output perturbation, can be used to obtain a better tradeoff for all convex-Lipschitz-bounded learning tasks. Since output perturbation is simple to implement, it means that our approach is potentially widely applicable in practice. We go on to apply it on the same medical data as used by Fredrikson et al. Encouragingly, we achieve accurate models even for $\varepsilon = 0.1$. In the last part of this paper, we study the impact of our improved differentially private mechanisms on model inversion attacks, a privacy attack introduced by Fredrikson et al. We observe that the improved tradeoff makes the resulting differentially private model more susceptible to inversion attacks. We analyze this phenomenon formally.