A5100676785- Natural Language Processing Techniques
- Topic Modeling
- Multimodal Machine Learning Applications
- Robot Manipulation and Learning
- Genomics and Phylogenetic Studies
- RNA and protein synthesis mechanisms
- Robotics and Sensor-Based Localization
- Robotic Path Planning Algorithms
- Human Pose and Action Recognition
- Advanced Image and Video Retrieval Techniques
- Biomedical Text Mining and Ontologies
- Reinforcement Learning in Robotics
- Advanced Neural Network Applications
- Advanced Vision and Imaging
- Speech and dialogue systems
- RNA modifications and cancer
- Algorithms and Data Compression
- Domain Adaptation and Few-Shot Learning
- Semantic Web and Ontologies
- Gene expression and cancer classification
- Text Readability and Simplification
- Machine Learning in Bioinformatics
- Advanced Sensor and Control Systems
- Image Processing Techniques and Applications
- IoT and Edge/Fog Computing
Shanghai University of Electric Power
2010-2025
Sorbonne Université
2025
Laboratoire Kastler Brossel
2024-2025
University of Washington
2012-2024
École Normale Supérieure
2024
Google (United States)
2023-2024
DeepMind (United Kingdom)
2024
Hunan University
2018-2024
Chinese Academy of Medical Sciences & Peking Union Medical College
2022-2024
Nanchang Institute of Technology
2024
With growing interest in Chinese Language Processing, numerous NLP tools (e.g., word segmenters, part-of-speech taggers, and parsers) for have been developed all over the world. However, since no large-scale bracketed corpora are available to public, these trained on with different segmentation criteria, tagsets bracketing guidelines, therefore, comparisons difficult. As a first step towards addressing this issue, we preparing large corpus late 1998. The two installments of corpus, 250...
Developing visual perception models for active agents and sensorimotor control in the physical world are cumbersome as existing algorithms too slow to efficiently learn real-time robots fragile costly. This has given rise learning-in-simulation which consequently casts a question on whether results transfer real-world. In this paper, we investigate developing real-world agents, propose Gibson Environment purpose, showcase set of perceptual tasks learned therein. is based upon virtualizing...
Large language models can encode a wealth of semantic knowledge about the world. Such could be extremely useful to robots aiming act upon high-level, temporally extended instructions expressed in natural language. However, significant weakness is that they lack real-world experience, which makes it difficult leverage them for decision making within given embodiment. For example, asking model describe how clean spill might result reasonable narrative, but may not applicable particular agent,...
Large language models excel at a wide range of complex tasks. However, enabling general inference in the real world, e.g., for robotics problems, raises challenge grounding. We propose embodied to directly incorporate real-world continuous sensor modalities into and thereby establish link between words percepts. Input our model are multi-modal sentences that interleave visual, state estimation, textual input encodings. train these encodings end-to-end, conjunction with pre-trained large...
Large language models (LLMs) trained on code-completion have been shown to be capable of synthesizing simple Python programs from docstrings [1]. We find that these code-writing LLMs can re-purposed write robot policy code, given natural commands. Specifically, code express functions or feedback loops process perception outputs (e.g., object detectors [2], [3]) and parameterize control primitive APIs. When provided as input several example commands (formatted comments) followed by...
By transferring knowledge from large, diverse, taskagnostic datasets, modern machine learning models can solve specific downstream tasks either zero-shot or with small taskspecific datasets to a high level of performance.While this capability has been demonstrated in other fields such as computer vision, natural language processing speech recognition, it remains be shown robotics, where the generalization capabilities are particularly critical due difficulty collecting real-world robotic...
Recent works have shown how the reasoning capabilities of Large Language Models (LLMs) can be applied to domains beyond natural language processing, such as planning and interaction for robots. These embodied problems require an agent understand many semantic aspects world: repertoire skills available, these influence world, changes world map back language. LLMs in environments need consider not just what do, but also when do them - answers that change over time response agent's own choices....
We study how vision-language models trained on Internet-scale data can be incorporated directly into end-to-end robotic control to boost generalization and enable emergent semantic reasoning. Our goal is a single model both learn map robot observations actions enjoy the benefits of large-scale pretraining language from web. To this end, we propose co-fine-tune state-of-the-art trajectory tasks, such as visual question answering. In contrast other approaches, simple, general recipe achieve...
Tracking 6-D poses of objects from videos provides rich information to a robot in performing different tasks such as manipulation and navigation. In this article, we formulate the object pose tracking problem Rao–Blackwellized particle filtering framework, where 3-D rotation translation an are decoupled. This factorization allows our approach, called PoseRBPF, efficiently estimate along with full distribution over rotation. is achieved by discretizing space fine-grained manner training...
Large language models (LLMs) have unlocked new capabilities of task planning from human instructions. However, prior attempts to apply LLMs real-world robotic tasks are limited by the lack grounding in surrounding scene. In this paper, we develop NLMap, an open-vocabulary and queryable scene representation address problem. NLMap serves as a framework gather integrate contextual information into LLM planners, allowing them see query available objects before generating context-conditioned...
The nose-to-brain pathway has been proven to be a shortcut for direct drug delivery the brain. However, whether and what extent nanoparticles can delivered through this passage is still awaiting validation with evidence. In study, transportation of tracked via fluorescence bioimaging strategies using nanoemulsions (NEs) as model carriers. Identification NEs in biological tissues based on → off signal switching new type environment-responsive embedded dyes, P2 P4, two conventional probes, DiR...
Land use and land cover (LULC) mapping in urban areas is one of the core applications remote sensing, it plays an important role modern planning management. Deep learning springing up field machine recently. By mimicking hierarchical structure human brain, deep can gradually extract features from lower level to higher level. The Belief Networks (DBN) model a widely investigated deployed architecture. It combines advantages unsupervised supervised archive good classification performance. This...
Long-read sequencing technologies have the potential to produce gold-standard de novo genome assemblies, but fully exploiting error-prone reads resolve repeats remains a challenge. Aggressive approaches repeat resolution often misassemblies, and conservative lead unnecessary fragmentation. We present HINGE, an assembler that seeks achieve optimal by distinguishing can be resolved given data from those cannot. This is accomplished adding "hinges" for constructing overlap graph where only...
We present iGibson 1.0, a novel simulation environment to develop robotic solutions for interactive tasks in large-scale realistic scenes. Our contains 15 fully home-sized scenes with 108 rooms populated rigid and articulated objects. The are replicas of real-world homes, distribution the layout objects aligned those real world. 1.0 integrates several key features facilitate study tasks: i) generation high-quality virtual sensor signals (RGB, depth, segmentation, LiDAR, flow so on), ii)...
The in vivo translocation of nanoemulsions (NEs) was tracked by imaging tools with an emphasis on the size effect. To guarantee accurate identification NEs vivo, water-quenching environment-responsive near-infrared fluorescent probes were used to label NEs. Imaging evidence confirmed prominent digestion gastrointestinal tract and oral absorption integral that survive enteric epithelia a size-dependent way. In general, reducing particle leads slowed vitro lipolysis digestion, prolonged...
We present Interactive Gibson Benchmark, the first comprehensive benchmark for training and evaluating Navigation: robot navigation strategies where physical interaction with objects is allowed even encouraged to accomplish a task. For example, can move if needed in order clear path leading goal location. Our comprises two novel elements: 1) new experimental setup, Environment (iGibson 0.5), which simulates high fidelity visuals of indoor scenes, dynamics common found these scenes; 2) set...
Contextual features always play an important role in Chinese word segmentation (CWS). Wordhood information, being one of the contextual features, is proved to be useful many conventional character-based segmenters. However, this feature receives less attention recent neural models and it also challenging design a framework that can properly integrate wordhood information from different measures existing frameworks. In paper, we therefore propose framework, WMSeg, which uses memory networks...
Inspired by research in psychology, we introduce a behavioral approach for visual navigation using topological maps.Our goal is to enable robot navigate from one location another, relying only on its observations and the map of environment.To this end, propose graph neural networks localizing agent map, decompose action space into primitive behaviors implemented as convolutional or recurrent networks.Using Gibson simulator Stanford 2D-3D-S dataset, verify that our outperforms relevant...
Humans can routinely follow a trajectory defined by list of images/landmarks. However, traditional robot navigation methods require accurate mapping the environment, localization, and planning. Moreover, these are sensitive to subtle changes in environment. In this letter, we propose PoliNet, deep visual model predictive control-policy learning method that perform while avoiding collisions with unseen objects on path. PoliNet takes as input 360° images from robot's current view outputs...
Recent research in embodied AI has been boosted by the use of simulation environments to develop and train robot learning approaches. However, skewed attention tasks that only require what robotics simulators can simulate: motion physical contact. We present iGibson 2.0, an open-source environment supports a more diverse set household through three key innovations. First, 2.0 object states, including temperature, wetness level, cleanliness toggled sliced necessary cover wider range tasks....
Large, high-capacity models trained on diverse datasets have shown remarkable successes efficiently tackling downstream applications. In domains from NLP to Computer Vision, this has led a consolidation of pretrained models, with general backbones serving as starting point for many Can such happen in robotics? Conventionally, robotic learning methods train separate model every application, robot, and even environment. we instead generalist X-robot policy that can be adapted new robots,...
Recent advances in robot learning have shown promise enabling robots to perform a variety of manipulation tasks and generalize novel scenarios.One the key contributing factors this progress is scale data used train models.To obtain large-scale datasets, prior approaches relied on either demonstrations requiring high human involvement or engineering-heavy autonomous collection schemes, both which being challenging scaling up space new skills needed for building generalist robots.To mitigate...