A5086941203- Evolutionary Algorithms and Applications
- Low-power high-performance VLSI design
- Geological Formations and Processes Exploration
- Paleontology and Stratigraphy of Fossils
- VLSI and FPGA Design Techniques
- Metaheuristic Optimization Algorithms Research
- Geology and Paleoclimatology Research
- Marine and environmental studies
- Analog and Mixed-Signal Circuit Design
- Evolution and Paleontology Studies
- Cellular Automata and Applications
- VLSI and Analog Circuit Testing
- Numerical Methods and Algorithms
- Advancements in Semiconductor Devices and Circuit Design
- Ancient and Medieval Archaeology Studies
- Neural Networks and Applications
- CRISPR and Genetic Engineering
- Viral Infectious Diseases and Gene Expression in Insects
- Geology and Environmental Impact Studies
- Advanced Memory and Neural Computing
- Paleontology and Evolutionary Biology
- Advanced Multi-Objective Optimization Algorithms
- Image and Signal Denoising Methods
- Radiation Effects in Electronics
- Cephalopods and Marine Biology
Brno University of Technology
2015-2024
Czech Academy of Sciences, Institute of Geonics
2013-2023
TU Dortmund University
2022
TU Wien
2019
VSB - Technical University of Ostrava
2006-2015
State Geological Institute of Dionýz Štúr
2015
Czech Academy of Sciences
2013-2014
Comenius University Bratislava
2012
Geological Survey of Canada
2012
St Charles Hospital
2001
Approximate circuits and approximate circuit design methodologies attracted a significant attention of researchers as well industry in recent years. In order to accelerate the system process support fair benchmarking approximation methods, we propose library adders multipliers called EvoApprox8b. This contains 430 non-dominated 8-bit created from 13 conventional 471 6 multipliers. These implementations were evolved by multi-objective Cartesian genetic programming. The EvoApprox8b provides...
Artificial neural networks (NN) have shown a significant promise in difficult tasks like image classification or speech recognition. Even well-optimized hardware implementations of digital NNs show power consumption. It is mainly due to non-uniform pipeline structures and inherent redundancy numerous arithmetic operations that be performed produce each single output vector. This paper provides methodology for the design power-efficient with uniform structure suitable implementation. An error...
In approximate computing, the requirement of perfect functional behavior can be relaxed because some applications are inherently error resilient. Approximate circuits, which fall into computing paradigm, designed in such a way that they do not fully implement logic given by specification and, hence, their accuracy exchanged for lower area, delay or power consumption. order to automate design process, we propose evolve digital circuits show minimal supplied amount resources. The is based on...
Improving the accuracy of a neural network (NN) usually requires using larger hardware that consumes more energy. However, error tolerance NNs and their applications allow approximate computing techniques to be applied reduce implementation costs. Given multiplication is most resource-intensive power-hungry operation in NNs, economical multipliers (AMs) can significantly In this article, we show AMs also improve NN by introducing noise. We consider two categories AMs: 1) deliberately...
The state-of-the-art approaches employ approximate computing to reduce the energy consumption of DNN hardware. Approximate DNNs then require extensive retraining afterwards recover from accuracy loss caused by use operations. However, complex does not scale well. In this paper, we demonstrate that efficient approximations can be introduced into computational path accelerators while completely avoided. ALWANN provides highly optimized implementations for custom low-power in which number units...
Libraries of approximate circuits are composed fully characterized digital that can be used as building blocks energy-efficient implementations hardware accelerators. They employed not only to speed up the accelerator development but also analyze how an responds introducing various operations. In this paper, we present a methodology automatically builds comprehensive libraries with desired properties. Target generated using Cartesian genetic programming. addition extending EvoApprox8b...
We present a novel method allowing one to approximate complex arithmetic circuits with formal guarantees on the approximation error. The integrates in unique way techniques for equivalence checking into search-based circuit optimisation algorithm. key idea of our approach is employ search strategy that drives towards promptly verifiable circuits. was implemented within ABC tool and extensively evaluated functional multipliers (with up 32-bit operands) adders 128-bit operands). Within few...
Approximate computing exploits the fact that many applications are inherently error resilient. In order to reduce power consumption, approximate circuits such as multipliers have been employed in these applications. However, most current based on ad hoc circuit structures and, for automated approximation methods, large efficient designs difficult find due increased search space. Moreover, existing design methods do not typically provide sufficient formal guarantees terms of if constructed....
A new FPGA implementation for adaptive median filters is proposed. Adaptive exhibit better filtering properties than standard filters; however, their cost higher. Proposed architecture was optimized throughput allowing 300 M pixels to be filtered per second. The best performance/cost ratio exhibits the filter which utilizes window 7x7 and can suppress shot noise with intensity up 60%. In addition filtering, also used as detectors of corrupted (detection statistics).
Gigantic rates of data production in the era Big Data, Internet Thing (IoT)/Internet Everything (IoE), and Cyber Physical Systems (CSP) pose incessantly escalating demands for massive processing, storage, transmission while continuously interacting with physical world under unpredictable, harsh, energy-/power-constrained scenarios. Therefore, such systems need to support not only high performance capabilities at tight power/energy envelop, but also be intelligent/cognitive, self-learning,...
Approximate circuits are digital that intentionally designed in such a way the specification is violated terms of functionality order to obtain some improvements power consumption, performance or area, comparison with fully functional circuits. To design approximate circuits, synthesis tools rely on availability procedure checking, whether synthesized meet and/or provides information about circuit quality. Compared traditional flow, nature involves replacing strict equivalence checking more...
Energy efficiency of hardware accelerators deep neural networks (DNN) can be improved by introducing approximate arithmetic circuits. In order to quantify the error introduced using these circuits and avoid expensive prototyping, a software emulator DNN accelerator is usually executed on CPU or GPU. However, this emulation typically two three orders magnitude slower than implementation running GPU operating with standard floating point instructions common libraries. The reason that there no...
In this paper, a new circuit architecture for image filter evolution is proposed. The evolvable system based on the implementation of search algorithm in PowerPC processor which available Xilinx Virtex II Pro Field Programmable Gate Arrays (FPGAs). Candidate filters are evaluated domain-specific virtual reconfigurable implemented using logic same FPGA. As enables to execute more sophisticated algorithms than an original solely circuit-based solution by Martinek and Sekanina, higher...
This paper deals with evolutionary design of approximate circuits. class circuits is characterized by relaxing the requirement on functional equivalence between specification and implementation in order to reduce area a chip or minimize energy consumption. We proposed CGP-based automated method which enables find good trade off key circuit parameters (functionality, power consumption). In particular, digital consisting elementary gates are addressed this paper. Experimental results provided...
The main issue connected with using approximate components such as multipliers in deep convolutional neural networks (CNN) during the design process is necessity to emulate them due lack of native support for operations modern CPUs and GPUs, which computationally expensive. To accelerate emulation CNNs on we propose TFApprox4IL, a software library supporting both symmetric well asymmetric quantization modes, 8xN bit emulated lookup tables, new type layer known depthwise convolution,...
Recently, power efficiency has become the most important parameter of many real circuits. At same time, a wide range applications capable tolerating imperfections spread out especially in multimedia. Approximate computing, an emerging paradigm, takes advantage relaxed functional requirements to make computer systems more efficient terms energy consumption, speed or complexity. As result, variety trade-offs between error and can be found. In this paper, design method based on multi-objective...