Article
Computer Science, Artificial Intelligence
Zhijun Zhang, Haotian He, Xianzhi Deng
Summary: A novel fuzzy recurrent neural network (FRNN) is proposed to resist internal error noises of robots, and it is designed and implemented on field-programmable gated array (FPGA).
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Information Systems
Jinsung Yoon, Neungyun Kim, Donghyun Lee, Su-Jung Lee, Gil-Ho Kwak, Tae-Hwan Kim
Summary: This paper proposes a resource-efficient keyword spotting system based on a convolutional neural network. The system utilizes a one-dimensional CNN for end-to-end keyword recognition, achieving high accuracy and fast processing speed. By binarizing the model and employing a dedicated engine, the system reduces resource usage without sacrificing performance. Experimental results show that the system achieves a processing latency of 22 ms and a spotting accuracy of 91.80% in a specific environment.
Article
Computer Science, Hardware & Architecture
Michaela Blott, Nicholas J. Fraser, Giulio Gambardella, Lisa Halder, Johannes Kath, Zachary Neveu, Yaman Umuroglu, Alina Vasilciuc, Miriam Leeser, Linda Doyle
Summary: Research provides theoretical evaluation of different CNN optimization techniques and hardware platform performance, finding channel pruning to be most effective and FPGA benefitting the most from quantization. Pruning and quantization are orthogonal and yield optimal design points when combined.
IEEE TRANSACTIONS ON COMPUTERS
(2021)
Article
Computer Science, Information Systems
Sansei Hori, Hakaru Tamukoh
Summary: This study proposes a hardware-oriented implementation method for a restricted Boltzmann machine without using random number generators. The method utilizes cut-off bits obtained from fixed-point binary arithmetic operations on digital hardware. The proposed method reduces hardware resource requirements and power consumption compared to other random number generators.
Article
Computer Science, Information Systems
Jinwon Kim, Jiho Kim, Tae-Hwan Kim
Summary: AERO is a resource-efficient reconfigurable inference processor designed for recurrent neural networks (RNN) of various types. It utilizes a versatile vector-processing unit (VPU) to achieve high resource efficiency by processing primitive vector operations and utilizing an approximation scheme for multiplication. The resource efficiency of AERO was found to be significantly higher than the previous state-of-the-art result, reaching 1.28 MOP/s/LUT.
Article
Computer Science, Information Systems
Zbigniew Hajduk, Grzegorz Rafal Dec
Summary: The paper presents a relatively simple method for implementing the hyperbolic tangent function on FPGAs using ordinary or Chebyshev polynomials. Various implementation versions with different polynomial degrees and intervals have been considered for both floating-point and fixed-point computations. The proposed method achieves high accuracy while maintaining reasonable resource utilization and calculation time. It can serve as an effective alternative to other methods like CORDIC and can be easily adapted for implementing other mathematical functions.
Article
Computer Science, Information Systems
Jiho Kim, Tae-Hwan Kim
Summary: This paper presents an efficient inference processor ROSETTA for RNNs, which achieves high-speed inference on devices with limited silicon resources and power supply by optimizing computational workload and power consumption.
IEEE TRANSACTIONS ON EMERGING TOPICS IN COMPUTING
(2023)
Article
Computer Science, Artificial Intelligence
Shuanglong Liu, Hongxiang Fan, Wayne Luk
Summary: Computing convolutional layers in the frequency domain using FFT can reduce computational complexity, but the frequent transformations between spatial and frequency domains hinder low-latency inference. To address this, a fully spectral CNN is proposed, which eliminates the transformations using a novel spectral-domain adaptive ReLU layer. Additionally, a customized hardware architecture is proposed to accelerate the fully spectral CNN inference on FPGA, achieving improved throughput compared to state-of-the-art implementations.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Ekaterina Govorkova, Ema Puljak, Thea Aarrestad, Thomas James, Vladimir Loncar, Maurizio Pierini, Adrian Alan Pol, Nicolo Ghielmetti, Maksymilian Graczyk, Sioni Summers, Jennifer Ngadiuba, Thong Q. Nguyen, Javier Duarte, Zhenbin Wu
Summary: The Large Hadron Collider generates a large number of collision events that need to be filtered in real-time. This paper demonstrates the deployment of unsupervised deep learning algorithms in a real-time event selection system to search for new physics signatures. The first-stage filter, implemented on custom electronics, significantly enhances the signal-over-background ratio.
NATURE MACHINE INTELLIGENCE
(2022)
Article
Computer Science, Artificial Intelligence
Jinming Lu, Chao Ni, Zhongfeng Wang
Summary: This paper introduces a method for efficient training of deep neural networks on resource-constrained platforms. A hardware-algorithm co-optimization approach is used to implement an efficient training accelerator on an FPGA. The proposed training scheme reduces computational complexity and memory accesses while maintaining accuracy. Experimental results demonstrate state-of-the-art accuracy across multiple models.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Engineering, Electrical & Electronic
Noriaki Kaneda, Chun-Yen Chuang, Ziyi Zhu, Amitkumar Mahadevan, Bob Farah, Keren Bergman, Doutje Van Veen, Vincent Houtsma
Summary: This article proposes a deep neural network based equalizer to tackle the intersymbol interference in high speed passive optical network (PON) links. The performance of the DNN based equalizer is found to be superior to the conventional equalizer in both back-back and through fiber experiments. To reduce hardware complexity, the article investigates embedded parallelization and classification output stage, and analyzes the impact of fixed-point resolution on hardware resource utilization.
JOURNAL OF LIGHTWAVE TECHNOLOGY
(2022)
Article
Engineering, Electrical & Electronic
Zhaopeng Xu, Shuangyu Dong, Jonathan H. Manton, William Shieh
Summary: With the rapid development of machine learning technologies, neural network (NN)-based equalizers have been proven to be efficient tools for dealing with nonlinear impairments in optical interconnects. However, the computational complexity (CC) of these equalizers remains a major concern. This paper proposes an NN-based multi-symbol equalization scheme inspired by multi-task learning to significantly reduce CC.
JOURNAL OF LIGHTWAVE TECHNOLOGY
(2022)
Article
Engineering, Electrical & Electronic
Mojan Javaheripi, Mohammad Samragh, Farinaz Koushanfar
Summary: Tensor decomposition is a promising method for implementing low-power and real-time neural network applications on resource-constrained embedded devices. The proposed AutoRank framework allows customization of neural network decomposition through cross-layer rank selection, incorporating both inference accuracy and platform specifications while minimizing engineering costs. This framework is hardware-aware and delivers high accuracy decomposed deep neural networks with low execution costs, with an automated API for compatibility with popular deep learning libraries.
IEEE JOURNAL ON EMERGING AND SELECTED TOPICS IN CIRCUITS AND SYSTEMS
(2021)
Article
Chemistry, Physical
Gulbahar Bilgic, Basak Ozturk, Sema Atasever, Mukerrem Sahin, Hakan Kaplan
Summary: In this study, a machine learning approach based on an artificial neural network model was developed to improve the hydrogen production system with a magnetic field effect. The model accurately predicted the effect of input parameters on hydrogen output and demonstrated strong predictive performance.
INTERNATIONAL JOURNAL OF HYDROGEN ENERGY
(2023)
Article
Engineering, Electrical & Electronic
Yashrajsinh Parmar, K. Sridharan
Summary: This study focuses on developing an area-time efficient VLSI architecture for a novel self-feedback Convolutional Neural Network (CNN), achieving high accuracy in object detection while significantly reducing on-chip memory requirement through an efficient systolic array architecture.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS
(2021)
Article
Thermodynamics
Karani Kurtulus, Ahmet Coskun, Shadan Ameen, Ceyhun Yilmaz, Ali Bolatturk
ENERGY CONVERSION AND MANAGEMENT
(2018)
Article
Energy & Fuels
Ceyhun Yilmaz
Article
Chemistry, Physical
Onder Kaska, Ceyhun Yilmaz, Onur Bor, Nehir Tokgoz
INTERNATIONAL JOURNAL OF HYDROGEN ENERGY
(2018)
Article
Chemistry, Physical
Ceyhun Yilmaz, Onder Kaska
INTERNATIONAL JOURNAL OF HYDROGEN ENERGY
(2018)
Article
Engineering, Electrical & Electronic
Murat Alcin, Ismail Koyuncu, Murat Tuna, Metin Varan, Ihsan Pehlivan
INTERNATIONAL JOURNAL OF CIRCUIT THEORY AND APPLICATIONS
(2019)
Article
Chemistry, Physical
Ceyhun Yilmaz
INTERNATIONAL JOURNAL OF HYDROGEN ENERGY
(2020)
Article
Computer Science, Hardware & Architecture
Murat Tuna, Murat Alcin, Ismail Koyuncu, Can Bulent Fidan, Ihsan Pehlivan
MICROPROCESSORS AND MICROSYSTEMS
(2019)
Article
Chemistry, Physical
Ismail Koyuncu, Ceyhun Yilmaz, Murat Alcin, Murat Tuna
INTERNATIONAL JOURNAL OF HYDROGEN ENERGY
(2020)
Article
Thermodynamics
Muhammed Arslan, Ceyhun Yilmaz
Summary: The use of biomass in energy conversions is crucial for a renewable and sustainable energy future. This study investigates the recovery of waste heat from exhaust gases in a biogas power plant using an integrated Organic Rankine Cycle. The results show that waste heat recovery improves the net power production and economic performance of the biogas power plant.
Article
Chemistry, Physical
Ceyhun Yilmaz, Ozan Sen
Summary: This study aims to optimize geothermal and solar-assisted sustainable energy and hydrogen production system using genetic algorithm. The integration of hydrogen as an energy storage unit brings sustainability to smart grid systems. The study uses artificial neural network (ANN) based genetic algorithm (GA) optimization technique to constantly study the system in the most suitable conditions, including unit product cost and power output. The overall energy and exergy efficiencies of the integrated system are found to be 4.97% and 16.0% respectively.
INTERNATIONAL JOURNAL OF HYDROGEN ENERGY
(2022)
Article
Green & Sustainable Science & Technology
Omer Faruk Guler, Ozan Sen, Ceyhun Yilmaz, Mehmet Kanoglu
Summary: This study models and analyzes the performance of a geothermal and solar-based multigeneration system, comparing it with alternative case studies. Three different models are developed, each using a different method to transfer heat energy. The study also considers the integration of a hydrogen production system to store excess energy. The models are investigated using geothermal and solar data from Afyonkarahisar in Turkey. The results show variations in exergy efficiencies and costs among the different models.
Article
Chemistry, Physical
Muhammed Arslan, Ceyhun Yilmaz
Summary: This study evaluates the biogas power production and green hydrogen potential from biomass. It integrates an Organic Rankine Cycle (ORC) to utilize waste exhaust gases. The power from the ORC is used for hydrogen production, H2S elimination, and excess electricity storage. Thermodynamic and thermoeconomic analyses and optimization of the Combined Heat and Power (CHP) system are performed. The results show improved efficiencies and costs compared to the existing power plant.
INTERNATIONAL JOURNAL OF HYDROGEN ENERGY
(2023)
Article
Computer Science, Interdisciplinary Applications
Ismail Koyuncu, Murat Alcin, Murat Tuna, Ihsan Pehlivan, Metin Varan, Sundarapandian Vaidyanathan
INTERNATIONAL JOURNAL OF COMPUTER APPLICATIONS IN TECHNOLOGY
(2019)
