A5030224453- Reinforcement Learning in Robotics
- Scientific Computing and Data Management
- Parallel Computing and Optimization Techniques
- Advanced Data Storage Technologies
- Cloud Computing and Resource Management
- Robotic Path Planning Algorithms
- Distributed and Parallel Computing Systems
- Evolutionary Algorithms and Applications
- Autonomous Vehicle Technology and Safety
- Context-Aware Activity Recognition Systems
- Laser-Plasma Interactions and Diagnostics
- Software System Performance and Reliability
- Algorithms and Data Compression
- Real-Time Systems Scheduling
- Astrophysics and Cosmic Phenomena
- Age of Information Optimization
- Embedded Systems Design Techniques
- Data Visualization and Analytics
- Computational Physics and Python Applications
- Evacuation and Crowd Dynamics
- Human-Automation Interaction and Safety
- Advanced biosensing and bioanalysis techniques
- AI-based Problem Solving and Planning
- Recommender Systems and Techniques
- Personal Information Management and User Behavior
Robert Bosch (Germany)
2016-2023
Robert Bosch (India)
2022
Robert Bosch (Taiwan)
2021
TU Dresden
2012-2017
Nvidia (United States)
2013-2017
Nvidia (United Kingdom)
2017
University of Lübeck
2009-2013
We present a particle-in-cell simulation of the relativistic Kelvin-Helmholtz Instability (KHI) that for first time delivers angularly resolved radiation spectra particle dynamics during formation KHI. This enables studying KHI with unprecedented spatial, angular and spectral resolution. Our results are great importance understanding astrophysical jet comparable plasma phenomena by relating motion observed in to its signature.
Solving robotic navigation tasks via reinforcement learning (RL) is challenging due to their sparse reward and long decision horizon nature. However, in many tasks, high-level (HL) task representations, like a rough floor plan, are available. Previous work has demonstrated efficient by hierarchal approaches consisting of path planning the HL representation using sub-goals derived from plan guide RL policy source task. these usually neglect complex dynamics sub-optimal sub-goal-reaching...
Inverse Reinforcement Learning (IRL) describes the problem of learning an unknown reward function a Markov Decision Process (MDP) from observed behavior agent. Since agent's originates in its policy and MDP policies depend on both stochastic system dynamics as well function, solution inverse is significantly influenced by both. Current IRL approaches assume that if transition model unknown, additional samples system's are accessible, or provides enough to solve accurately. These assumptions...
Programming of high performance computing systems has become more complex over time. Several layers parallelism need to be exploited efficiently utilize the available resources. To support application developers and analysts we propose a technique for identifying most critical optimization targets in distributed heterogeneous applications. We have developed CASITA, tool which uses an execution trace knowledge about programming models MPI, OpenMP CUDA as well their hierarchy among each other...
The use of accelerators in heterogeneous systems is an established approach designing petascale applications. Today, Compute Unified Device Architecture (CUDA) offers a rich programming interface for GPU but requires developers to incorporate several layers parallelism on both the CPU and GPU. From this increasing program complexity emerges need sophisticated performance tools. This work contributes by analyzing hybrid MPI-CUDA programs properties based wait states, such as critical path,...
In this paper we present a concept for integrating multimedia-enriched learning objects (MELOs) within unifying technological framework schools. First, discuss the general role of child development and learning. We then outline architecture proposed networked environment multimedia (NEMO). Afterwards, give examples to describe how use MELOs could enhance experience primary school children while about dolphins pedagogical benefits our approach.
Utilizing accelerators in heterogeneous systems is an established approach for designing peta-scale applications. Today, CUDA offers a rich programming interface GPU but requires developers to incorporate several layers of parallelism on both CPU and GPU. From this increasing program complexity emerges the need sophisticated performance tools. This work contributes by analyzing hybrid MPI-CUDA programs their critical path, property proven effectively identify application bottlenecks. We...
One of the challenges for developer highly-parallel MPI applications running on distributed high performance computing systems is to understand complex behavior their applications. It requires identify inefficiencies, and optimize them such that communication waiting times can be reduced. This task only accomplished with help elaborated tools provide insight into details application using an automatic analysis or intuitive visualization approach. While first target a specific problem domain,...
More and more computationally intensive scientific applications make use of hardware accelerators like general purpose graphics processing units (GPGPUs). Compared to software development for typical multi-core processors their programming is fairly complex needs specific optimizations utilize the full computing power. To achieve high performance, critical parts a program have be identified optimized. This paper proposes an approach performance analysis CUDA kernel source code regions, which...
Emerging new technologies in plasma simulations allow tracking billions of particles while computing their radiative spectra.We present a visualization the relativistic Kelvin-Helmholtz Instability from simulation performed with fully particle-in-cell code PIConGPU powered by 18,000 GPUs on USA's fastest supercomputer Titan [1].
Summary Event‐based performance monitoring and analysis are effective means when tuning parallel applications for optimal resource usage. In this article, we address the data capacity challenge that arises applying tracing methodology to large‐scale long execution times. Existing approaches use static, pre‐defined event filters reduce a manageable size. contrast, propose self‐guided automatically adapt an application's runtime behaviour therefore, do not require any previous knowledge or...
In this paper we address the problem of massive event trace sizes, one most urgent challenges in performance analysis large-scale parallel applications. Reducing sizes during application runtime decreases slow down, eliminates measurement bias, and cuts down stress on underlying file system. Previous approaches use static filters to decrease size, which relies preceding knowledge about or, otherwise, delivers poor results. contrast, propose that automatically adapt an application's behavior...
Maximum Causal Entropy (MCE) Inverse Optimal Control (IOC) has become an effective tool for modelling human behaviour in many control tasks. Its advantage over classic techniques estimating policies is the transferability of inferred objectives: Behaviour can be predicted variations task by policy computation using a relaxed optimality criterion. However, exact inference often computationally intractable problems with imperfect state observation. In this work, we present model class that...