3

Click here to load reader

PERSONAL INFORMATION Nitin Satpute 21, 2017 · Curriculum vitae Nitin Satpute 4) A Flexible Scalable Hardware Architecture for Radial Basis Function Neural Networks (RBFNN): Implementation

Embed Size (px)

Citation preview

Page 1: PERSONAL INFORMATION Nitin Satpute 21, 2017 · Curriculum vitae Nitin Satpute 4) A Flexible Scalable Hardware Architecture for Radial Basis Function Neural Networks (RBFNN): Implementation

Curriculum vitae

PERSONAL INFORMATION Nitin Satpute

Campus de Rabanales, Universidad de Cordoba (CIF Q1418001B), 14071 Cordoba (Spain)

(+34) 684222048 (+91) 8105352958

[email protected]

https://nitinrsatpute.wordpress.com/ https://www.hipeac.net/~nitinsatpute/

Sex Male | Date of birth 12 May 1990 | Nationality Indian

WORK EXPERIENCE

12 Jun 2017–Present ResearcherUniversity of Cordoba (http://www.uco.es/), Cordoba (Spain)

GPU Implementation of Region Growing Methods: I am involved in the development and implementation of high performance parallel algorithms for region growing on GPUs. A paper on "Evaluation of GPU Region Growing Methods on NVIDIA GPUs" is accepted in JAI 2017 (III JornadasAndaluzas de Informática 2017). I am writing a study paper on performance comparison and evaluation of region growing algorithms on different platforms. This study explores the impact of various parameters such as seed selection, connectivity (neighborhood) criteria, threshold etc on performance and accuracy of region growing implementation. Based on the study, the optimized region growing algorithm will be implemented on the high performance GPU platform (such as K40) ormobile platforms (such as TX1) (Environments: Linux, C++, CUDA, & Python, GTX 780 GPU).

1 Jan 2013–4 Jun 2017 Previous Work Experience1. Indian Institute of Science (IISc) Bangalore, India (www.iisc.ernet.in/) 2. University of Siena, Italy (http://en.unisi.it/) and 3. Visvesvaraya National Institute of Technology (VNIT) Nagpur, India (http://vnit.ac.in/)

Research Areas: High Performance and Scientific Computing, GPU Computing, Convolutional Deep Neural Network (CNN), Medical Image Analysis, Mathematical Statistics, Artificial Neural Networks, Computer Architecture, H.264 & H.265 Video Decoders

Projects: 1) Matrix Multiplication Performance Characterization on GPUs with a Single Point (Poster inProgramming and tUning Massively Parallel Systems - PUMPS 2015 at BSC, Spain): The goals of Matrix Multiplication (MM) performance characterization on GPUs are to a) Distinguish performance ofSGEMM and DGEMM on different GPU platforms, algorithms and other parameters b) Identify characteristics of MM on different platforms (HW/SW) for clusters of similar performance values c) Compare the performance with other platforms. We introduce criteria for classifying performance by using a technique that we call “Linear Deviation Point". The aim of this technique is to identify a single point to characterize e.g. weak scaling curve or strong scaling curve. Based on such point we further compare performance of several GPU platforms and derive conclusions.

2) Comparing and Evaluating GPU Platforms with a Single Point (Presented poster and published a research paper in ACACES - 2015 Summer School): We have seen from the literature that it's not immediate to compare performance results from different published research papers.We propose a methodology to summarize typical curves used to indicate the performance of a given computing platform. We present performance of MM on a GPU platform using a single point. We form clusters of points with similar performance values. We have found three clusters for the SGEMM, SGEMM & DGEMM, and DGEMM. We compare and evaluate platforms in these clusters based on the underlying hardware platform, precision, algorithm, library, etc. (Environment - Linux, C, C++, Cuda)

3) Convolutional Deep Neural Network (CNN) for Breast Cancer Detection: I have implemented convolutional, activation (ReLU and Sigmoid), pooling, fully connected and softmax layers. I have developed mathematical models for robust feature learning. It includes learning from adversarial examples and making classifiers perform robustly when confronted with such hard positive/negative examples. The parameters responsible for feature learning are number of each of the mentioned layers, selection of the activation functions, pooling parameters (max or mean, stride depth, pool size) and the type and size of the filters. The flow starts by choosing number of parameters randomly and initializing them with random numbers. Then the experimental and mathematical analysis is done to update the filters. I have implemented back-propagation algorithm (stochastic gradient descent (SGD)) for error minimization. (Environment - Linux, Python and MATLAB)

21/9/17 © European Union, 2002-2017 | http://europass.cedefop.europa.eu Page 1 / 3

Page 2: PERSONAL INFORMATION Nitin Satpute 21, 2017 · Curriculum vitae Nitin Satpute 4) A Flexible Scalable Hardware Architecture for Radial Basis Function Neural Networks (RBFNN): Implementation

Curriculum vitae Nitin Satpute

4) A Flexible Scalable Hardware Architecture for Radial Basis Function Neural Networks (RBFNN): Implementation of ANN on ASIC developed by researchers is both expensive and time consuming considering the complexity of recent embedded platforms. In view of this, hardware solution provided by the network of reconfigurable HyperCells is suggested and they are known to provide high performance by decreasing the execution time in a wide range of applications. (Environment - Linux, Matlab, Python and Bluespec System Verilog (BSV))

5) Bulk synchronous parallel computing (BSPC) on H.265 video decoder: BSPC provides a robust model for parallel computation and a programming framework which facilitates the development of portable algorithms and performance prediction. It is capable of providing a cost model to design, analyse and optimize massively parallel algorithms for various applications. (Environment - Linux, Python and BSP Programming) (https://www.youtube.com/watch?v=6m4Tt4305_E&feature=youtu.be)

6) FPGA realization of H.264 Video Decoder: This work was initiated during my period of stay at IISc as an Intern from Jan, 2013 to Jun, 2013 and continued further as a Project Associate. H.264 Video Decoder has been implemented using Bluespec System Verilog. The implementation supports baseline profile, capable of decoding High Definition (HD) video. (Environment - Linux, and BSV)

EDUCATION AND TRAINING

1 Aug 2011–30 Jun 2013 Master of Engineering in Embedded Systems (CGPA : 7.92/10)Birla Institute of Technology & Science Pilani, India (www.bits-pilani.ac.in/)

Projects: 1) Basic CPU design using Verilog HDL and implementation on FPGA: Implemented 8 bit CPU using Verilog which supports instructions for addition, multiplication, load and store.

2) Tabu Search based implementation of object tracking using Joint Colour Texture Histogram: A new robust methodology has been proposed which uses Tabu search algorithm along with joint colour texture histogram to track a moving object efficiently. (Environment - Windows and Matlab)

1 Aug 2007–30 Jun 2011 Bachelor of Engineering (Percentage: 74.5% with 95/100 maximummarks in Engineering Mathematics)Yeshwantrao Chavan College of Engineering Nagpur, India (www.ycce.edu/)

Specialization: Electronics and Telecommunication Engineering

Project: Segmentation & Classification of MRI Brain Images using Texture Features (Worked under Asst. Prof. Yogita Dubey and Prof. M. M. Mushrif): Segmentation and classification of MRI brain images as normal or abnormal are performed using K Nearest Neighbour classification algorithm with large number of input test images for better accuracy. Dealt with the some of the Mathematical Statistical concepts related to segmentation and clustering algorithms such as Distributions, Bayes theorem, Entropy, Mean, Mode, Median etc. (Environment - Windows and Matlab)

PERSONAL SKILLS

Organisational / managerial skills a) Adaptive to multicultural environments, gained through my work and educational experiences b) Team Spirit and Organisational or Managerial skills: Gained various experiences such as leadership and team management while working on a research projects and organising various workshops/training programs.

Job-related skills Computer Vision, Mathematical Statistics, Image & Video Processing, Embedded Systems, ComputerArchitecture, High Performance & Scientific Computing, GPU Computing, CUDA and OpenMP

Digital competence Operating System, Programming and Tools: Linux, Windows, Xilinx ISE and SDK, C, C++, OOPS, CUDA, Cilk, MATLAB, Python, Verilog, and Bluespec System Verilog

Other skills Trekking, Travelling, Cooking, Listening Music, Gymnasium, Playing Chess and Volleyball

21/9/17 © European Union, 2002-2017 | http://europass.cedefop.europa.eu Page 2 / 3

Page 3: PERSONAL INFORMATION Nitin Satpute 21, 2017 · Curriculum vitae Nitin Satpute 4) A Flexible Scalable Hardware Architecture for Radial Basis Function Neural Networks (RBFNN): Implementation

Curriculum vitae Nitin Satpute

ADDITIONAL INFORMATION

Publications 1) Nitin Satpute, Juan G omez-Luna, and Joaqu ın Olivares, "Evaluation of GPU Region Growing Methods on NVIDIA GPUs" in JAI (III Jornadas Andaluzas de Informática) 2017, Malaga, Spain.

2) Nitin Satpute, Roberto Giorgi, "Comparing and Evaluating GPU Platforms with a Single Point" in theEleventh International Summer School on Advanced Computer Architecture and Compilation for High-Performance and Embedded Systems - ACACES 2015.

3) Mahnaz Mohammadi, Nitin Satpute, Rohit Ronge, Jayesh Chandiramani, S K Nandy, Aamir Raihan, Tanmay Verma, Ranjani Narayan and Sukumar Bhattacharya, "A Flexible Scalable HardwareArchitecture for Radial Basis Function Neural Networks" in the International Conference on VLSI Design and Embedded Systems, 2015.

4) Bharath Kumar Koora, Nitin Satpute and Akshay Adiga, "Tabu Search based implementation of object tracking using Joint Colour Texture Histogram" in the International Conference on Industrial andInformation Systems, IIT Chennai, IEEE Explore, August 6, 2012, pp. 1-6.

5) Anisha Shriram, Nikhil Dhabekar, Murtaja Hussain, Pranjali Jumle and Nitin Satpute, "Segmentation & Classification of MRI Brain Images using Texture Features" in the International Journal of Machine Intelligence & Applications:2011.

6) Nitin Satpute, "Matrix Multiplication Performance Characterization on GPUs with a Single Point" presented a poster in Programming and tUning Massively Parallel Systems - PUMPS 2015 at Barcelona Supercomputing Center (BSC), Spain.

Honours and awards 1) Awarded Marie Skodowska-Curie grant ("living allowance" of 3035,36€/month and a "mobility allowance" of 600€/month) from the project High Performance Soft-tissue Navigation (HIPERNAV - H2020-MSCA-ITN-2016) in an Innovative Training Network (ITN)

2) Conferred funding from Ministry of Electronics and Information Technology (MeitY), India Rs.37800 per Month (Year 2016-2017) and funded acceptance from PUMPS 2015, Spain and ACACES 2015, Italy Summer Schools

3) Received GATE scholarship of Rs 8000 per month (Aug 2011 - July 2013) and full fee waiver of approx. Rs 2,40,000 for securing one of the top five ranks in the State Engineering Entrance Examination (Aug, 2007 - June, 2011)

4) Percentage: 86.83% in XII with 93% marks in Maths and 88.26% in X with 91.33% marks in Maths and awarded with Rs. 5000 cash from the Central Government of India (Aug, 2005)

Other Activities 1. Organized training programs on "Play with Arduino" on Sept, 2016 and "Advanced Arduino with Matlab & Raspberry Pi" on March, 2017 at Prof. Ram Meghe Institute of Technology & Research, Badnera, Maharashtra 444701

2. Attended a) MIT GSW at Novotel Hyderabad Convention Centre (Mar, 2016) b) Deep Learning Training program presented by Deep Learning Institute, nVIDIA & hosted by GPU Center of Excellence, IIT Bombay on Dec 05, 2016 c) "The 2015 LOFAR Surveys Meeting" held at Leiden, Netherlands (Sept, 2015) d) "AXIOM Face to Face Meet" held at BSC, Spain (Jun, 2015) e) Summer training program at 3-Base Repair Depot, Indian Air Force, Chandigarh (May-June, 2010)

3. Assisted Prof. Donald Reay from Heriot-Watt University, UK in conducting Faculty Development Program on DSP for Educators at IIIT Bangalore, VNIT Nagpur and NIT Patna (Mar & Sep, 2016)

4. Delivered a talk on Technology & Opportunities (Mar, 2016) and conducted a workshop on "Image and Video Processing using Linux, Python and Opencv on Raspberry Pi" at Sir MVIT College, Bangalore (Apr, 2016)

References Prof. (Dr.) Joaquín Olivares Bueno (Computer Architecture and Electronics Department, University of Cordoba, Spain) - [email protected] (+34) 957 212062

Prof. (Dr.) Juan Gómez Luna (Computer Architecture and Electronics Department, University of Cordoba, Spain) - [email protected] (+34) 646 099 449

Prof. (Dr.) Donald Reay (Institute of Sensors, Signals and Systems, Heriot-Watt University, UK) - [email protected] +44(0)1314513359

Prof. (Dr.) Prasanna L. Zade (Dept. of ECE, YCCE, India) - [email protected]

21/9/17 © European Union, 2002-2017 | http://europass.cedefop.europa.eu Page 3 / 3