hello world!

i am an r&d engineer at synopsys inc. i did my master's degree in data science at indiana university, bloomington and bachelor's degree in computer engineering from nirma university, ahmedabad. i have five years of research experience in the field of artificial intelligence and machine learning. my projects involve computer vision, medical imaging, remote sensing, signal processing, nuclear physics, parallel computing, and genomics.

i am very passionate about body building and fitness.

research interest

applications of machine learning and deep learning in almost any domain.

news

• jun 2023: started working at synopsys inc. as an r&d engineer.
• may 2023: received excellence award in research during my masters.
• may 2023: graduated with a masters of science degree in data science from indiana university.
• jan 2023: paper on predicting noncoding insertion and deletion in human genome got published.
• dec 2022: paper on predicting promoter-centered chromatin interactions got published.
• oct 2022: started my internship at synopsys inc. as a machine learning engineer.
• jul 2022: bagged 1st prize in summer research showcase for work on computer vision in biology.

read more

• oct 2021: started research assistant position at indiana university.
• sep 2021: our paper on plasma disruption accepted in plasma physics & controlled fusion.
• aug 2021: started my master's in data science at indiana university, bloomington.
• jun 2021: gave a lecture on bitcoin and cyptocurrencies at hsbc.
• jan 2021: passed the tensorflow developer certification exam.
• oct 2020: our paper on prostate segmentation accepted in pattern recognition & image analysis.
• sep 2020: bagged the hsbc hero award for our exceptional work at hsbc during covid-19 pandemic.
• aug 2020: got certified as aws machine learning specialist, developer, and solutions architect.
• apr 2020: gave a talk on super-resolution using deep learning at nirma university, ahmedabad.
• oct 2019: graduated from nirma university with bachelor's in computer engineering.
• sep 2019: received the most innovative idea award for our work on authenticating users from electrocardiogram signals and deep learning methods.
• jul 2019: joined hsbc software development, india as a software engineer.
• apr 2019: conducted nvidia dli workshop on topics of computer vision, dl for multiple data types, and cuda programming at mahindra école centrale, hyderabad.
• apr 2019: presented our poster on prostate segmentation at nvidia gpu technology conference (gtc), san jose.
• sep 2018: conducted a lecture on the basics of machine learning at nirma university.
• jul 2015: started my bachelor's in computer engineering at nirma university.
• mar 1997: deployed on earth.

publications

journal

• tivan-indel: a computational framework for annotating and predicting noncoding regulatory small insertions and deletions.
  aman agarwal, fengdi zhao, yuchao jiang, and li chen.
  bioinformatics, 2023, doi 10.1093/bioinformatics/btad060.

• deepphic: predicting promoter-centered chromatin interactions using a novel deep learning approach.
  aman agarwal, li chen.
  bioinformatics, 2022, doi 10.1093/bioinformatics/btac801.

• deep sequence to sequence learning-based prediction of major disruptions in aditya tokamak.
  aman agarwal, aditya mishra, priyanka sharma, swati jain, raju daniel, sutapa ranjan, ranjana manchanda, joydeep ghosh, and rakesh tanna.
  plasma physics and controlled fusion, 2021, doi 10.1088/1361-6587/ac234c.

• dilated volumetric network: an enhanced fully convolutional network for volumetric prostate segmentation from magnetic resonance imaging.
  aman agarwal, aditya mishra, madhushree basavarajaiah, priyanka sharma, and sudeep tanwar.
  pattern recognition and image analysis, 31, 228–239 (2021), doi 10.1134/S1054661821020024.

poster

• volumetric prostate segmentation from mri using fully-connected neural network.
  aman agarwal, aditya mishra, and priyanka sharma.
  presented at nvidia gpu technology conference (gtc), san jose, 2019.

open-source projects

detecting tree throws in lidar images [code]

• yolov5 library was customized for geotiff images to detect points of interests.
• algorithm was trained such that it can work for images at different resolutions and from different regions.

sentence simplication using transformers [code] [paper]

• rewriting complex sentences in simpler forms for easier understanding while maintaining its original meaning.
• gpt-2 and bert transformer models were trained on wikilarge dataset and evaluated on popular metric for the task.
• the best model scored an average sari score of 46.8 on the test set, beating the sota by 3.49.

tracking polymer tips in plant cells [code]

• tracking microscopic intracellular movements in plants using deep learning methods.
• a u-net model was trained on simulated images and evaluated on actual ones due to the absence of labeled data.
• this approach can reduce 100s of hours of manual labeling work to just a few minutes.

wildfire detection from satellite images [code]

• designed an algorithm to segment wildfires from real-time satellite images using vision transformers.
• outperformed the segmentation results of the published baseline model by nearly 5% points.

using ecg for biometric authentication [code] [blog]

• ecg signals from smartwatch are passed to a siamese network hosted on aws.
• the network converts the ecg signals to frequency spectrogram and verifies the user.
• ecg is unique for an individual and is very promising for this task. reference

object detection in satellite images [code] [blog]

• a deep network to detect ships in oceans from real-time satellite images using yolov3.
• trained a customized model on darknet and hosted it on aws to download the latest satellite images from planet labs, make predictions, & send the detected object coordinates to the user.

3d prostate segmentation of mr images using fcnn [paper] [project page]

• our enhanced v-net model outperformed the results of the baseline in the promise12 challenge.
• the model was enhanced by tweaking the architecture, adding dilation, and deep supervision. we improved the accuracy by 6%.

predicting the dynamics of tokamak discharge [paper]
(department of atomic energy, india)

• the aim of the project was to anticipate the phenomenon of major disruption in plasma confinement for aditya tokamak.
• we were able to anticipate the disruption of plasma 12 ms prior to the actual disruption (4 ms earlier than the state-of-the-art models).
• input features included the readings of various diagnostics like plasma current, mirnov oscillations, loop voltage, bolo meter readings, and many other.

our model predicting the disruption in tokamak plasma in real time.

speech emotion recognition [code]

• prediction of human emotions from raw audio using iemocap database.
• bidirectional lstm was used along with local attention mechanism to focus on the part of speech which influence the emotion more.
• the architecture was trained on nvidia k80 system and gave results comparable to the state-of-the-art models.

autonomous car [paper] [code]

• a self-driving rc car that can maneuver itself on an indoor, hand-made track.
• convolutional neural network was used to classify the direction of car from dashcam images.
• the model was deployed on raspberry pi for real-time predictions.

---

aman agarwal