Deployment of deep neural networks for applications that require very high throughput or extremely low latency is a severe computational challenge, further exacerbated by inefficiencies in mapping the computation to h...
详细信息
ISBN:
(纸本)9781728199023
Deployment of deep neural networks for applications that require very high throughput or extremely low latency is a severe computational challenge, further exacerbated by inefficiencies in mapping the computation to hardware. We present a novel method for designing neural network topologies that directly map to a highly efficient FPGA implementation. By exploiting the equivalence of artificial neurons with quantized inputs/outputs and truth tables, we can train quantized neural networks that can be directly converted to a netlist of truth tables, and subsequently deployed as a highly pipelinable, massively parallel FPGA circuit. However, the neural network topology requires careful consideration since the hardware cost of truth tables grows exponentially with neuron fan-in. To obtain smaller networks where the whole netlist can be placed-and-routed onto a single FPGA, we derive a fan-in driven hardware cost model to guide topology design, and combine high sparsity with few-bit activation quantization to limit the neuron fan-in. We evaluate our approach on two tasks with very high intrinsic throughput requirements in high-energy physics and network intrusion detection. We show that the combination of sparsity and few-bit activation quantization results in high-speed circuits with small logic depth and low LUT cost, demonstrating competitive accuracy with less than 15 ns of inference latency and throughput in the hundreds of millions of inferences per second.
Nowadays a lot of control systems are based on safety PLC (safety programmable logic controllers). the most of these PLC's are certified for SIL3 application (Safety integrity level 3). When components of this saf...
详细信息
ISBN:
(纸本)9781728143811
Nowadays a lot of control systems are based on safety PLC (safety programmable logic controllers). the most of these PLC's are certified for SIL3 application (Safety integrity level 3). When components of this safety PLC are not used right then resulting SIL is lower. But when more components are used for realization of safety function properly with application diagnostics, then resulting SIL of SF (safety function) can be higher than SIL of used components. this paper deals with possibilities of various SRCS architectures with application diagnostics to reach required SIL.
this paper presents the development of a Human-Robot Interface (HRI) that uses an RGB-D sensor to facilitate industrial robot programming. the main aims and functionalities of the HRI are described and subsequently em...
详细信息
this paper presents the development of a Human-Robot Interface (HRI) that uses an RGB-D sensor to facilitate industrial robot programming. the main aims and functionalities of the HRI are described and subsequently emphasis is placed on two novel user interactions methods through which the operator's movements and gestures are captured and used to position the joints of a 6 DOF articulated robot arm. In particular, during the first user interaction method the operator's torso, shoulder and elbow are used to determine the position of the first three robot joints, while the operator's gestures are used to determine the position of the remaining three joints. During the second user interaction method, a combination of the operator's movements and gestures are used to select any desirable robot joint and then modify its angle by a user-configurable step. Both interaction methods also allow the operator to control robot speed and gripper position. the HRI is complemented by a virtual model of the robotic arm, created in Unity3D, to evaluate online programming capabilities. Results from using the two developed user interaction methods to position the robot in pre-defined poses are presented and discussed to highlight strengths and weaknesses of each method. Finally, the obtained conclusions are presented and suggestions for future work are made. (C) 2021 the Authors. Published by Elsevier Ltd.
Considering high complexity and abstraction of object-oriented programming concepts, there is a need for innovative solution, which may serve as mean for learning these concepts. the aim of this paper is the descripti...
详细信息
ISBN:
(纸本)9781728143811
Considering high complexity and abstraction of object-oriented programming concepts, there is a need for innovative solution, which may serve as mean for learning these concepts. the aim of this paper is the description of developed interactive educational application for object-oriented-programming. We decided to use Unity, which is a cross-platform 3D game engine. We made our interactive part of the application as a drag and drop game, in which the user creates solution for given problem using knowledge, which were presented to him in the introduction part of the application. After the user finishes the challenge, he needs to pass the quiz. In each level there are several questions for particular concepts, on which the user has to answer. In this part the user can verify his gained knowledge. We have collected feedback from the questionnaire.
Analytic tableau systems for the family of non-deterministic semantics are introduced. these are based on tableaux for many-valued logics using sets-as-signs DNF representations. Karnaugh maps illustrate the construct...
详细信息
ISBN:
(纸本)9783030860592;9783030860585
Analytic tableau systems for the family of non-deterministic semantics are introduced. these are based on tableaux for many-valued logics using sets-as-signs DNF representations. Karnaugh maps illustrate the construction of tableau rules. In contrast to classical many-valued tableaux, we add a rule called sign intersection. Soundness and completeness are shown. As an example demonstrates, some tableau systems would be incomplete without sign intersection. there is a correspondence to well-studied canonical calculi based on sequent systems: Tableau systems can be translated into canonical calculi, but not vice-versa (structural rules are missing on the tableau side).
Flexible robotics will be a major enabling technology for the application of robot-based automation in other than traditionally suitable automotive or electronics production with high volumes. Increased demand for fle...
详细信息
Flexible robotics will be a major enabling technology for the application of robot-based automation in other than traditionally suitable automotive or electronics production with high volumes. Increased demand for flexibility due to individualized production typical for most SMEs require an increased level of flexibility - also for robots that should be able to learn as well as provide an increased level of autonomy due to improved skills and extended reasoning capabilities. this publication tries to find out if novel ANN methodology that is able to process 3D surface data is applicable to generalize process knowledge in a one shot learning by demonstration situation in order to be able to execute tasks on similar but geometrically unequal objects in future settings. the methodology generalizes not on symbolic or trajectory level but on surface geometry level and was applied to a simple geometric object on lab scale. the algorithms introduced are applicable to more complex objects with practical relevance. (C) 2021 the Authors. Published by Elsevier Ltd.
Surface-hardening process of steel materials by robot laser technologies can involve the challenge of modeling the determining process parameters through non-conventional tools in order to evaluate the quality of the ...
详细信息
Surface-hardening process of steel materials by robot laser technologies can involve the challenge of modeling the determining process parameters through non-conventional tools in order to evaluate the quality of the heat treatment. In the current study a new method based on fractal geometry, used to determine the microstructural properties of laser hardened steels manufactured by anthropomorphic robots, is presented. the assumptions were that the microstructure of laser hardened steel can be studied as a complex structural geometry and the modeling of the analyzed complex geometries can be made through genetic programming for prediction purposes. the effect of process parameters and their joint combination on the final microstructures geometry of the heat treated steel was investigated. In particular, the influence of temperature, laser beam velocity, and impact angle were studied since they were showed in a preliminary study to be the process parameters that most significantly influenced the quality of the heat treated steel. the developed model reached a precision of the prediction equal to 98.59 %. (C) 2021 the Authors. Published by Elsevier Ltd.
Nowadays, our lives are surrounded by various information, where large pieces of information are uncertain even inconsistent. And in user-oriented applications, users’ preferences should also be taken into account. T...
详细信息
Common criticisms of state-of-the-art machine learning include poor generalisation, a lack of interpretability, and a need for large amounts of training data. We survey recent work in inductive logicprogramming (ILP)...
详细信息
the amount and variety of data that we produce every day pose a constant challenge for meaningful information processing. While knowledge graphs have gained a considerable attention in the recent years, due to their f...
详细信息
ISBN:
(纸本)9789897584749
the amount and variety of data that we produce every day pose a constant challenge for meaningful information processing. While knowledge graphs have gained a considerable attention in the recent years, due to their flexible and universal knowledge representation, they lack the mechanisms facilitating knowledge processing. In this paper, we propose an information system called extended knowledge graphs (EKG) that augments the concept of knowledge graphs with procedural attachments. We put forward the requirements and assumption of the EKG structure and present a categorisation of supported reasoning tasks.
暂无评论