A5101670036- Transportation and Mobility Innovations
- Transportation Planning and Optimization
- Sharing Economy and Platforms
- Traffic control and management
- Circular RNAs in diseases
- Cell Adhesion Molecules Research
- Cancer-related gene regulation
- Electric Vehicles and Infrastructure
- Smart Parking Systems Research
- Autonomous Vehicle Technology and Safety
- Advanced Bandit Algorithms Research
- Prostate Cancer Treatment and Research
- Advanced Neural Network Applications
- RNA Research and Splicing
- Angiogenesis and VEGF in Cancer
- Software Engineering Research
- Blockchain Technology Applications and Security
- Machine Learning in Materials Science
- Optimization and Search Problems
- Platelet Disorders and Treatments
- Genomics, phytochemicals, and oxidative stress
- Wnt/β-catenin signaling in development and cancer
- Ginger and Zingiberaceae research
- Sparse and Compressive Sensing Techniques
- Cannabis and Cannabinoid Research
Sixth Affiliated Hospital of Sun Yat-sen University
2022-2023
Sun Yat-sen University
2022-2023
Menlo School
2023
BC Platforms (Finland)
2022
Meta (United States)
2022
Third Xiangya Hospital
2020
Central South University
2020
Didi Chuxing (China)
2019
Northeast Electric Power University
2017
Huazhong Agricultural University
2014
Recent works on ride-sharing order dispatching have highlighted the importance of taking into account both spatial and temporal dynamics in process for improving transportation system efficiency. At same time, deep reinforcement learning has advanced to point where it achieves superhuman performance a number fields. In this work, we propose based solution conduct large scale online A/B tests DiDi's ride-dispatching platform show that proposed method significant improvement total driver...
Ride dispatching is a central operation task on ride-sharing platform to continuously match drivers trip-requesting passengers. In this work, we model the ride problem as Markov Decision Process and propose learning solutions based deep Q-networks with action search optimize policy for platforms. We train evaluate agents challenging decision using real-world spatio-temporal trip data from DiDi platform. A large-scale system typically supports many geographical locations diverse demand-supply...
Order dispatching is instrumental to the marketplace engine of a large-scale ride-hailing platform, such as DiDi which continuously matches passenger trip requests drivers at scale tens millions per day. Because dynamic and stochastic nature supply demand in this context, order-dispatching problem challenging solve for an optimal solution. Added complexity are considerations system response time, reliability, multiple objectives. In paper, we describe how our approach optimization has...
How to optimally dispatch orders vehicles and how trade off between immediate future returns are fundamental questions for a typical ride-hailing platform. We model as large-scale parallel ranking problem study the joint decision-making task of order dispatching fleet management in online platforms. This brings unique challenges following four aspects. First, facilitate huge number act learn efficiently robustly, we treat each region cell an agent build multi-agent reinforcement learning...
Order dispatching and driver repositioning (also known as fleet management) in the face of spatially temporally varying supply demand are central to a ride-sharing platform marketplace. Hand-crafting heuristic solutions that account for dynamics these resource allocation problems is difficult, may be better handled by an end-to-end machine learning method. Previous works have explored methods problem from high-level perspective, where method responsible either drivers or orders, further...
Large ride-hailing platforms, such as DiDi, Uber and Lyft, connect tens of thousands vehicles in a city to millions ride demands throughout the day, providing great promises for improving transportation efficiency through tasks order dispatching vehicle repositioning. Existing studies, however, usually consider two simplified settings that hardly address complex interactions between two, real-time fluctuations supply demand, necessary coordinations due large-scale nature problem. In this...
Ride hailing has become prevailing. Central in ride platforms is taxi order dispatching which involves recommending a suitable driver for each order. Previous works use pure combinatorial optimization solutions dispatching, suffer practice due to complex dynamics of demand and supply temporal dependency among decisions. Recent studies try adopt data-driven method into hoping knowledge from history data would help overcome these challenges. Among attempts, adoption reinforcement learning...
The low rate of adoption by human users often hinders AI algorithms from achieving their intended efficiency gains. This is particularly true for that prioritize system-wide objectives because they can create misalignment incentives and cause confusion among potential users. We provide one the first large-scale field studies on algorithm aversion leveraging an algorithmic recommendation rollout a large ridesharing platform. identify contextual experience herding as two important factors...
AI algorithms often cannot realize their intended efficiency gains because of low adoption by human users. We uncover various factors that explain ridesharing drivers’ aversion to an algorithm designed help them make better location choices. By leveraging algorithmic recommendation rollout on a large platform, we find drivers are more averse the when they face higher cost implementing its instructions, experience suggests greater opportunity following algorithm, and peers’ actions contradict...
For both the traditional street-hailing taxi industry and recently emerged on-line ride-hailing, it has been a major challenge to improve ride-hailing marketplace efficiency due spatio-temporal imbalance between supply demand, among other factors. Despite numerous approaches using pricing dispatch strategies, they usually optimize or separately. In this paper, we show that these two processes are in fact intrinsically interrelated. Motivated by observation, make an attempt simultaneously...
Gastric cancer (GC) is a fatal with unclear pathogenesis. In this study, we explored the function and potential mechanisms of intercellular adhesion molecule 2 (ICAM2) in development advancement GC.Quantitative real-time polymerase chain reaction (qRT-PCR) Western blotting were performed to quantify ICAM2 expression harvested GC tissues cultured cell lines. Immunohistochemical analyses conducted on tissue microarray explore its implication prognosis patients. vitro experiments carried out...
A disintegrin and metalloproteinase with thrombospondin motifs 16 (ADAMTS16) has been reported to be involved in the pathogenesis of solid cancers. However, its role gastric cancer (GC) is unclear. In this study, ADAMTS16 was investigated. The effects on cell migration, invasion, proliferation were investigated by functional experiments vivo vitro. Downstream signal pathways confirmed using bioinformatics analysis, co-immunoprecipitation, immunofluorescence. Meanwhile, qRT-PCR, western blot,...
In this study, a scalable and real-time dispatching algorithm based on reinforcement learning is proposed for the first time, deployed in large scale. Current methods ridehailing platforms are dominantly myopic or rule-based non-myopic approaches. Reinforcement enables policies that informed of historical data able to employ learned information optimize returns expected future trajectories. Previous studies field yielded promising results, yet have left room further improvements terms...
Ladinin-1 (LAD1), an anchoring filament protein, has been associated with several cancer types, including cancers of the colon, lungs, and breast. However, it is still unclear how why LAD1 causes gastric (GC).Multiple in vitro vivo, functional gains loss experiments were carried out current study to confirm function LAD1. Mass spectrometry was used find proteins that interact Immunoprecipitation analyses revealed mechanism involved promoting aggressiveness.The results overexpressed GC...
Circular RNAs (circRNAs) are a novel class of noncoding RNAs. Increasing evidence indicates that circRNAs play an important role in the occurrence and development tumors. However, circRNA hsa_circ_0044556 progression colorectal cancer (CRC) remains unclear. First, we searched for differentially expressed using microarray paired CRC adjacent normal tissues. The was screened out from existing Gene Expression Omnibus database our microarray. clinical significance expression level patients then...
Existing approaches for vehicle repositioning on large-scale ride-hailing platforms either ignore the spatial-temporal mismatch between supply and demand in real-time or overlook long-term balance of system. To account both, we propose a lookahead policy this paper, which is novel approach to idle vehicles from both dynamic system performance perspective. Our method consists two parts; first part utilizes linear programming (LP) formulate nonstationary as time-varying, <inline-formula...
Transformers have enabled breakthroughs in NLP and computer vision, recently began to show promising performance trajectory prediction for Autonomous Vehicle (AV). How efficiently model the interactive relationships between ego agent other road dynamic objects remains challenging standard attention module. In this work we propose a general Transformer-like architectural module MnM network equipped with novel masked goal conditioning training procedures AV prediction. The resulted model,...
In this demo, we will present a simulation-based human-computer interaction of deep reinforcement learning in action on order dispatching and driver repositioning for ride-sharing. Specifically, demonstrate through several specially designed domains how use to train agents (drivers) have longer optimization horizon cooperate achieve higher objective values collectively.
The NeurIPS 2020 Procgen Competition was designed as a centralized benchmark with clearly defined tasks for measuring Sample Efficiency and Generalization in Reinforcement Learning. remains one of the most fundamental challenges deep reinforcement learning, yet we do not have enough benchmarks to measure progress community on We present design Learning which can help by doing end evaluation training rollout phases thousands user submitted code bases scalable way. top already existing...