A5101514892- Medical Imaging Techniques and Applications
- Advanced X-ray and CT Imaging
- Radiation Dose and Imaging
- Natural Language Processing Techniques
- Supercapacitor Materials and Fabrication
- Nanomaterials for catalytic reactions
- Protein Hydrolysis and Bioactive Peptides
- Advanced Photocatalysis Techniques
- Mathematics Education and Teaching Techniques
- Machine Learning and Data Classification
- Machine Learning in Bioinformatics
- Advanced Nanomaterials in Catalysis
- Topic Modeling
- Gamma-ray bursts and supernovae
- Child and Adolescent Psychosocial and Emotional Development
- Advanced biosensing and bioanalysis techniques
- Reinforcement Learning in Robotics
- Orthopedic Infections and Treatments
- Seismic Imaging and Inversion Techniques
- Adsorption and biosorption for pollutant removal
- Pulsars and Gravitational Waves Research
- Auditing, Earnings Management, Governance
- Infectious Diseases and Tuberculosis
- Human Pose and Action Recognition
- Membrane Separation Technologies
Tsinghua University
2019-2025
Yili Normal University
2024
China Agricultural University
2022-2024
Qufu Normal University
2024
Donghua University
2022-2023
State Power Investment Corporation (China)
2023
China Academy of Space Technology
2020
LVPYRP, a novel ACE inhibitory peptide identified from foxtail millet protein hydrolysates, can maintain stability under different food processing conditions, which is conducive to industrial production.
Limited angle reconstruction is a typical ill-posed problem in computed tomography (CT). Given incomplete projection data, images reconstructed by conventional analytical algorithms and iterative methods suffer from severe structural distortions artifacts. In this paper, we proposed self-augmented multi-stage deep-learning network (Sam's Net) for end-to-end of limited CT. With the merit alternating minimization technique, Sam's Net integrates self-constraints into cross-domain optimization...
In real-world sequential decision making tasks like autonomous driving, robotics, and healthcare, learning from observed state-action trajectories is critical for imitation, classification, clustering. For example, self-driving cars must replicate human driving behaviors, while robots healthcare systems benefit modeling sequences, whether or not they come expert data. Existing trajectory encoding methods often focus on specific rely reward signals, limiting their ability to generalize across...
Limited-view computed tomography (CT) presents significant potential for reducing radiation exposure and expediting the scanning process. While deep learning (DL) methods have exhibited promising results in mitigating streaking artifacts caused by a reduced number of projection views, their generalization remains challenging. In this work, we proposed DL-driven alternative Bayesian reconstruction method (DLBayesian) that efficiently integrates data-driven priors data consistency constraints....
Sailor2 is a family of cutting-edge multilingual language models for South-East Asian (SEA) languages, available in 1B, 8B, and 20B sizes to suit diverse applications. Building on Qwen2.5, undergoes continuous pre-training 500B tokens (400B SEA-specific 100B replay tokens) support 13 SEA languages while retaining proficiency Chinese English. Sailor2-20B model achieves 50-50 win rate against GPT-4o across languages. We also deliver comprehensive cookbook how develop the an efficient manner,...
Abstract Multi-segment static computed tomography (MS-staticCT) is a generalized and efficient configuration of CT systems, achieving high temporal resolution imaging by sequentially firing x-ray sources, instead rotation. However, it contains numerous geometric parameters. Due to the dense arrangement both sources detectors within their respective configurations, there are some coupled illumination relationships where simultaneously illuminate multiple detectors. To address these...
Simultaneous recovery of energy and carbon from recalcitrant wastewater has attracted ever-growing interest for water management. However, the existing technologies to break down pollutants are mainly chemical intensive. Here, a novel hydrothermal reaction amended with activated (AC) was demonstrated enable an unprecedented 99.5% removal exemplar difficult-to-degrade contaminant, polyvinyl alcohol (PVA), wastewater. Meanwhile, easy-separated hydrochar (C6H7.08O0.99) abundance unsaturated...
With the growing popularity of environmental, social, and governance (ESG), ESG performance is becoming increasingly important in investors’ decisions about firms. Capital market liberalization brings more sophisticated mature foreign investors who are interested corporate performance. We investigate whether capital improves disclosure using Shanghai-Hong Kong Stock Connect Shenzhen-Hong as exogenous shocks. By compiling a comprehensive dataset Chinese A-share listed firms from 2006 to 2019...
Sparse-view computed tomography (CT) has great potential in reducing radiation dose and accelerating the scan process. Although deep learning (DL) methods have exhibited promising results mitigating streaking artifacts caused by very few projections, their generalization remains a challenge. In this work, we proposed DL-driven alternative Bayesian reconstruction method that efficiently integrates data-driven priors data consistency constraints. This methodology involves two stages: universal...
Children's peer relationship plays a crucial role in the development process of children, and having good communication skills can help children shape personality qualities promote children's positivity. The formation optimistic emotions allows to carry out social cognition skills, thereby enabling sociality. In relationships, on ability now has more significant impact, problems based reality domestic foreign research, need be further solved, so impact is worth in-depth development....
Human alignment in large language models (LLMs) is an active area of research. A recent groundbreaking work, direct preference optimization (DPO), has greatly simplified the process from past work reinforcement learning human feedback (RLHF) by bypassing reward stage RLHF. DPO, after training, provides implicit model. In this we make a novel observation that model can itself be used bootstrapping fashion to further align LLM. Our approach use rewards current LLM construct dataset, which then...
In the wave of digital era, chip plays a key role in promoting whole information technology revolution with its core position development world science and technology. As cornerstone computing data processing, chips has promoted progress entire field. The chip’s manufacturing is to determining performance power chip, fault detection production process an important part product’s quality cost. Therefore, under background Moore’s law approaching physical limit, how ensure yield while...
Many problems in Reinforcement Learning (RL) seek an optimal policy with large discrete multidimensional yet unordered action spaces; these include randomized allocation of resources such as placements multiple security and emergency response units, etc. A challenge this setting is that the underlying space categorical (discrete unordered) large, for which existing RL methods do not perform well. Moreover, require validity realized (allocation); constraint often difficult to express...
In reasoning tasks, even a minor error can cascade into inaccurate results, leading to suboptimal performance of large language models in such domains. Earlier fine-tuning approaches sought mitigate this by leveraging more precise supervisory signals from human labeling, larger models, or self-sampling, although at high cost. Conversely, we develop method that avoids external resources, relying instead on introducing perturbations the input. Our training approach randomly masks certain...
Large language models (LLMs) have demonstrated impressive task-solving capabilities, achieved through either prompting techniques or system designs. However, concerns arisen regarding their proficiency in planning tasks, as they often struggle to generate valid plans. This paper investigates the impact of fine-tuning on LLMs' capabilities. Our findings indicate that LLMs can achieve good performance substantial (thousands specific examples) fine-tuning. is associated with significant...
Interior tomography is a typical strategy for radiation dose reduction in computed tomography, where only certain region-of-interest (ROI) scanned. However, given the truncated projection data, ROI reconstruction by conventional analytical algorithms may suffer from severe cupping artifacts. In this paper, we proposed new extraction-based deep learning method of interior tomography. Our approach works dual domains sinogram-domain network (SDNet) estimates contribution exterior region to and...
Abstract Activated carbon (AC) adsorption is a prevalent method for printing and dyeing wastewater (PDW) treatment, but restricted by active site depletion. The regeneration thermal desorption or chemical oxidation liable to destroy structure as well consume composition, leading few service cycles. Herein, we establish sustainable hydrothermal methylene blue (MB) exhausted AC. absorbed MB was converted its pyrrolic-N-containing hydrochar attached on regenerated AC , supporting 11...