Jiaming Hu

Yiyou Sun, Xinyang Han, Weichen Zhang et al.

Recent AI systems have achieved strong results on a wide range of benchmarks, yet these gains have not translated into economically meaningful deployment across many professional domains. We argue that this gap is largely an evaluation problem: widely used benchmarks lack sustained performance measurement on real and economically valuable workflows. This paper introduces Agents' Last Exam (ALE), a benchmark designed to evaluate AI agents on long-horizon, economically valuable, real-world tasks with verifiable outcomes. Developed in collaboration with 250+ industry experts, ALE covers non-physical industries defined with reference to O*NET / SOC 2018 (the U.S. federal occupational taxonomy). It is organized around a task taxonomy with 55 subfields grouped into 13 industry clusters covering 1K+ tasks. Current results show that the hardest tier remains far from saturated: across mainstream harness and backbone configurations, the average full pass rate is 2.6%. ALE is designed as a living benchmark: its task pool grows continuously as new workflows and industries are onboarded. More broadly, ALE is intended not merely as another leaderboard, but as an instrument for closing the gap between benchmark success and GDP-relevant impact.

Jiawei Wang, Dingyou Wang, Jiaming Hu et al.

A deep understanding of kinematic structures and movable components is essential for enabling robots to manipulate objects and model their own articulated forms. Such understanding is captured through articulated objects, which are essential for tasks such as physical simulation, motion planning, and policy learning. However, creating these models, particularly for objects with high degrees of freedom (DoF), remains a significant challenge. Existing methods typically rely on motion sequences or strong assumptions from hand-curated datasets, which hinders scalability. In this paper, we introduce Kinematify, an automated framework that synthesizes articulated objects directly from arbitrary RGB images or textual descriptions. Our method addresses two core challenges: (i) inferring kinematic topologies for high-DoF objects and (ii) estimating joint parameters from static geometry. To achieve this, we combine MCTS search for structural inference with geometry-driven optimization for joint reasoning, producing physically consistent and functionally valid descriptions. We evaluate Kinematify on diverse inputs from both synthetic and real-world environments, demonstrating improvements in registration and kinematic topology accuracy over prior work.

Rui Tian, Mingfei Gao, Mingze Xu et al.

We introduce UniGen, a unified multimodal large language model (MLLM) capable of image understanding and generation. We study the full training pipeline of UniGen from a data-centric perspective, including multi-stage pre-training, supervised fine-tuning, and direct preference optimization. More importantly, we propose a new Chain-of-Thought Verification (CoT-V) strategy for test-time scaling, which significantly boosts UniGen's image generation quality using a simple Best-of-N test-time strategy. Specifically, CoT-V enables UniGen to act as both image generator and verifier at test time, assessing the semantic alignment between a text prompt and its generated image in a step-by-step CoT manner. Trained entirely on open-source datasets across all stages, UniGen achieves state-of-the-art performance on a range of image understanding and generation benchmarks, with a final score of 0.78 on GenEval and 85.19 on DPG-Bench. Through extensive ablation studies, our work provides actionable insights and addresses key challenges in the full life cycle of building unified MLLMs, contributing meaningful directions to the future research.

Jiaming Hu, Jiamu Bai, Haoyu Wang et al.

Reinforcement learning from human feedback (RLHF) has been popular for aligning text-to-image (T2I) diffusion models with human preferences. As a mainstream branch of RLHF, Direct Preference Optimization (DPO) offers a computationally efficient alternative that avoids explicit reward modeling and has been widely adopted in diffusion alignment. However, existing preference-based methods for diffusion alignment still rely on reward-induced preference signals and typically assume that human preferences can be adequately modeled by the Bradley--Terry (BT) model, which may fail to capture the full complexity of human preferences. In this paper, we formulate diffusion alignment from a game-theoretic perspective. We propose Diffusion Nash Preference Optimization (Diff.-NPO), an intuitive general preference framework for diffusion alignment. Diff.-NPO encourages the current policy to play against itself to achieve self improvement and lead to a better alignment. Empirically, we demonstrate the effectiveness of Diff.-NPO on the text-to-image generation task via various metrics. Diff.-NPO consistently outperforms existing preference-based diffusion alignment methods.

Roman Bachmann, Oğuzhan Fatih Kar, David Mizrahi et al.

Current multimodal and multitask foundation models like 4M or UnifiedIO show promising results, but in practice their out-of-the-box abilities to accept diverse inputs and perform diverse tasks are limited by the (usually rather small) number of modalities and tasks they are trained on. In this paper, we expand upon the capabilities of them by training a single model on tens of highly diverse modalities and by performing co-training on large-scale multimodal datasets and text corpora. This includes training on several semantic and geometric modalities, feature maps from recent state of the art models like DINOv2 and ImageBind, pseudo labels of specialist models like SAM and 4DHumans, and a range of new modalities that allow for novel ways to interact with the model and steer the generation, for example image metadata or color palettes. A crucial step in this process is performing discrete tokenization on various modalities, whether they are image-like, neural network feature maps, vectors, structured data like instance segmentation or human poses, or data that can be represented as text. Through this, we expand on the out-of-the-box capabilities of multimodal models and specifically show the possibility of training one model to solve at least 3x more tasks/modalities than existing ones and doing so without a loss in performance. This enables more fine-grained and controllable multimodal generation capabilities and allows us to study the distillation of models trained on diverse data and objectives into a unified model. We successfully scale the training to a three billion parameter model using tens of modalities and different datasets. The resulting models and training code are open sourced at 4m.epfl.ch.

Shaokai Ye, Vasileios Saveris, Yihao Qian et al.

Image captioning is one of the most fundamental tasks in computer vision. Owing to its open-ended nature, it has received significant attention in the era of multimodal large language models (MLLMs). In pursuit of ever more detailed and accurate captions, recent work has increasingly turned to reinforcement learning (RL). However, existing captioning-RL methods and evaluation metrics often emphasize a narrow notion of caption quality, inducing trade-offs across core dimensions of captioning. For example, utility-oriented objectives can encourage noisy, hallucinated, or overlong captions that improve downstream question answering while harming fluency, whereas arena-style objectives can favor fluent but generic descriptions with limited usefulness. To address this, we propose a more balanced RL framework that jointly optimizes utility-aware correctness, reference coverage, and linguistic quality. In order to effectively optimize the resulting continuous multi-objective reward formulation, we apply GDPO-style reward-decoupled normalization to continuous-valued captioning rewards and show that it improves performance over vanilla GRPO. Additionally, we introduce length-conditional reward masking, yielding a more suitable length penalty for captioning. Across LLaVA-1.5-7B and Qwen2.5-VL 3B and 7B base models, our method consistently improves caption quality, with peak gains of +13.6 DCScore, +9.0 CaptionQA, and +29.0 CapArena across different models.

Yeping Jin, Jiaming Hu, Ioannis Ch. Paschalidis

Large Language Models (LLMs) tend to respond correctly to prompts that align to the data they were trained and fine-tuned on. Yet, small shifts in wording, format, or language can trigger surprisingly large failures, especially on multi-step reasoning problems. To address this problem, we propose a Distributionally Robust Token Optimization (DRTO) approach, which combines token-level Reinforcement Learning from Human Feedback (RLHF) with Distributionally Robust Optimization (DRO). DRTO bounds worst case token-wise rewards by constructing an f-divergence ambiguity set over a loss minibatch, leading to a theoretical robustness. Empirically, DRTO enhances consistency under distribution shifts in mathematical reasoning benchmarks, achieving 9.17\% improvement on GSM8K and 2.49% improvement on MathQA.

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.

Rui Yang, Jiaming Hu, Jian-Qing Zheng et al.

Artificial intelligence (AI) holds significant promise for enhancing intraoperative perception and decision-making in telesurgery, where physical separation impairs sensory feedback and control. Despite advances in medical AI and surgical robotics, conventional electronic AI architectures remain fundamentally constrained by the compounded latency from serial processing of inference and communication. This limitation is especially critical in latency-sensitive procedures such as endovascular interventions, where delays over 200 ms can compromise real-time AI reliability and patient safety. Here, we introduce an Optical Computation-in-Communication (OCiC) framework that reduces end-to-end latency significantly by performing AI inference concurrently with optical communication. OCiC integrates Optical Remote Computing Units (ORCUs) directly into the optical communication pathway, with each ORCU experimentally achieving up to 69 tera-operations per second per channel through spectrally efficient two-dimensional photonic convolution. The system maintains ultrahigh inference fidelity within 0.1% of CPU/GPU baselines on classification and coronary angiography segmentation, while intrinsically mitigating cumulative error propagation, a longstanding barrier to deep optical network scalability. We validated the robustness of OCiC through outdoor dark fibre deployments, confirming consistent and stable performance across varying environmental conditions. When scaled globally, OCiC transforms long-haul fibre infrastructure into a distributed photonic AI fabric with exascale potential, enabling reliable, low-latency telesurgery across distances up to 10,000 km and opening a new optical frontier for distributed medical intelligence.

Jiaming Hu, Haoyu Wang, Debarghya Mukherjee et al.

Jailbreak attacks pose a serious challenge to the safe deployment of large language models (LLMs). We introduce CCFC (Core & Core-Full-Core), a dual-track, prompt-level defense framework designed to mitigate LLMs' vulnerabilities from prompt injection and structure-aware jailbreak attacks. CCFC operates by first isolating the semantic core of a user query via few-shot prompting, and then evaluating the query using two complementary tracks: a core-only track to ignore adversarial distractions (e.g., toxic suffixes or prefix injections), and a core-full-core (CFC) track to disrupt the structural patterns exploited by gradient-based or edit-based attacks. The final response is selected based on a safety consistency check across both tracks, ensuring robustness without compromising on response quality. We demonstrate that CCFC cuts attack success rates by 50-75% versus state-of-the-art defenses against strong adversaries (e.g., DeepInception, GCG), without sacrificing fidelity on benign queries. Our method consistently outperforms state-of-the-art prompt-level defenses, offering a practical and effective solution for safer LLM deployment.

Jiaming Hu, Debarghya Mukherjee, Ioannis Ch. Paschalidis

In many real-world applications, ensuring the robustness and stability of deep neural networks (DNNs) is crucial, particularly for image classification tasks that encounter various input perturbations. While data augmentation techniques have been widely adopted to enhance the resilience of a trained model against such perturbations, there remains significant room for improvement in robustness against corrupted data and adversarial attacks simultaneously. To address this challenge, we introduce DRO-Augment, a novel framework that integrates Wasserstein Distributionally Robust Optimization (W-DRO) with various data augmentation strategies to improve the robustness of the models significantly across a broad spectrum of corruptions. Our method outperforms existing augmentation methods under severe data perturbations and adversarial attack scenarios while maintaining the accuracy on the clean datasets on a range of benchmark datasets, including but not limited to CIFAR-10-C, CIFAR-100-C, MNIST, and Fashion-MNIST. On the theoretical side, we establish novel generalization error bounds for neural networks trained using a computationally efficient, variation-regularized loss function closely related to the W-DRO problem.

Aayush Naik, Priyam Parashar, Jiaming Hu et al.

The World Robot Summit (WRS) 2020 Assembly Challenge is designed to allow teams to demonstrate how one can build flexible, robust systems for assembly of machined objects. We present our approach to assembly based on integration of machine vision, robust planning and execution using behavior trees and a hierarchy of recovery strategies to ensure robust operation. Our system was selected for the WRS 2020 Assembly Challenge finals based on robust performance in the qualifying rounds. We present the systems approach adopted for the challenge.

Priyam Parashar, Aayush Naik, Jiaming Hu et al.

The World Robotics Challenge (2018 & 2020) was designed to challenge teams to design systems that are easy to adapt to new tasks and to ensure robust operation in a semi-structured environment. We present a layered strategy to transform missions into tasks and actions and provide a set of strategies to address simple and complex failures. We propose a model for characterizing failures using this model and discuss repairs. Simple failures are by far the most common in our WRC system and we also present how we repaired them.

Jiaming Hu, Hongyi Ling, Priyam Parashar et al.

In the robotic industry, specular and textureless metallic components are ubiquitous. The 6D pose estimation of such objects with only a monocular RGB camera is difficult because of the absence of rich texture features. Furthermore, the appearance of specularity heavily depends on the camera viewpoint and environmental light conditions making traditional methods, like template matching, fail. In the last 30 years, pose estimation of the specular object has been a consistent challenge, and most related works require massive knowledge modeling effort for light setups, environment, or the object surface. On the other hand, recent works exhibit the feasibility of 6D pose estimation on a monocular camera with convolutional neural networks(CNNs) however they mostly use opaque objects for evaluation. This paper provides a data-driven solution to estimate the 6D pose of specular objects for grasping them, proposes a cost function for handling symmetry, and demonstrates experimental results showing the system's feasibility.