Deepayan Roy

Deepayan Roy

Data Science Graduate
India

About Me

Deepayan Roy is a Data Science graduate with research experience in machine learning, deep learning, computer vision, and audio processing. He has worked on projects involving noise cancellation, road safety detection, i…

Experience

Research Intern

University of Salento, Lecce, Italy
Mar 2023 - Jul 2023 · 4 months

Contributed to AI-based electronic system design for phased arrays and smart RF radio front ends, optimizing system architecture.
Utilized advanced data analysis techniques, including deep learning, to enhance system performance.
Collaborated cross-functionally to integrate AI solutions, improving system capabilities.
Applied data science and machine learning for data-driven system optimization.
Conducted statistical analysis to identify data patterns and inform decision-making.
Managed data preprocessing, transformation, and visualization for clear communication of results.

Masters’ Thesis – AI-based electronic system to drive phased arrays and smart RF radio front-ends

Vellore Institute of Technology (VIT), Amravati
Sep 2021 - May 2023 · 1 year 8 months

Studied previous models such as deep visual audio denoising (DVAD), AUDSCITY (Audio Denoising by Adaptive Social CosparsITY), and convolutional denoising autoencoders (CDAEs) that have ability to separate noise from real audio.
Analyzed the real time audio dataset with spectrogram and MFCC (Mel-frequency cepstral coefficients).
Created a CNN model using UNet and ResNet deep learning algorithm.
Calculated the accuracy and model loss using mean square error.
Transformed the stored neural network from an h5 file into an h5 lite file using TensorFlow lite or micro python.
Transferred the new h5 lite file into the MAX78000 microcontroller.
Developed a noise cancellation system which can reduce background noise from real time audio.

Research Project – Automatic Image and Video frame Text detection and conversion using Tesseract

Vellore Institute of Technology, Amravati
Sep 2022 - Dec 2022 · 3 months

Used Tesseract OCR to identify text from pictures and video frames.
Included features such as no typing, raw data editing, speedy translation, and memory utilization.
Converted detected text to audio to assist visually impaired people in hearing the information they need.

Research Project – Machine Learning Model to Ensure Road Safety/Reduce Accidents

Vellore Institute of Technology, Amravati
Aug 2022 - Nov 2022 · 3 months

Collected real time footage with the help of drone and CCTV cameras.
Trained the model with Pascal VOC (visual object classes) and Microsoft COCO (common objects in context) data sets.
Used the YOLO algorithm to detect the type of the vehicle.
Used MobileNet SSD to calculate the speed of the vehicle.
Compared the precision, mAP, and F1 score of the model with other models such as SSD-512, DSS-513, and STDN-513.

Research Project – NECLAT CLOSED: A Vertical Algorithm for Mining Frequent Closed Itemsets

Vellore Institute of Technology, Amravati
Jul 2022 - Oct 2022 · 3 months

Reduced the runtime and memory requirements of frequent item-set mining tasks.
Provided a lossless and concise collection of all frequent item sets by using frequent closed item sets.
Applied a fast-mining algorithm for frequent closed itemsets called NECLAT-CLOSED based on the concept and methods of vertical database format.
Compared with leading algorithms and found NECLAT-CLOSED performance better in terms of runtime and memory usage.

Research Project – Image Captioning with Visual Attention for Visually Impaired People

Vellore Institute of Technology, Amravati
Mar 2022 - Jul 2022 · 4 months

Developed a new text based visual attention (TBVA) model that automatically focuses on specific salient objects by eliminating irrelevant information when given previously generated text.
Suggested a multimodal recurrent neural network architecture as the foundation for the end-to-end caption generating mechanism.
Increased performance while lowering the number of parameters by employing the transposed weight-sharing approach.
Verified the efficiency of the model using MS COCO and Flickr30k.

Research Project – Hand written Digit Recognition

Vellore Institute of Technology, Amravati
Dec 2021 - Feb 2022 · 2 months

Applied deep learning techniques using Python and its libraries and frameworks.
Recognized handwritten digits from 0 to 9.
Used the MNIST dataset and achieved outputs by modeling a neural network trained over a dataset containing images of digits by the CNN model.

Research Project – Heart Failure Analysis and Prediction

Vellore Institute of Technology, Amravati
Sep 2021 - Dec 2021 · 3 months

Analyzed the correlation between factors including age, sex, sodium creatinine, sodium serum, creatinine phosphokinase, diabetes, anemia, high blood pressure, ejection fraction, platelets, smoking, time, and death event.

Skills

Big Data Data Mining Python SQL Database Data Science Data Representation Linear Algebra Machine Learning Algorithms Pytorch R Programming Language TensorFlow Deep Learning MongoDB Signal Processing R Programming SQL MATLAB MAX78000 Microcontroller Microsoft Excel Microsoft Word TensorFlow Lite Micro Python UNet ResNet CNN YOLO MobileNet SSD Tesseract OCR MFCC Spectrogram Machine Learning Information Retrieval Time Series Analysis Regression Analysis Image Processing Artificial Intelligence Reinforcement Learning Statistical Analysis Data Preprocessing Data Transformation
Report this Profile?