Aviral Chharia

I am a Master's Student at Carnegie Mellon University. I am broadly interested in the fields of Artificial Intelligence, Machine Learning, Deep Learning, Computer Vision and Biomedical Informatics.

During my undergrad, I was primarily advised by Prof. Vinay Kumar and Prof. Rahul Upadhyay at the Thapar Artificial Intelligence Lab and the Thapar AI & Biomedical Imaging Research Labs respectively. I was also fortunate enough to collaborate with Prof. Neeraj Kumar on some amazing research projects. I have been working closely with Prof. Apurva Narayan at the UBC Intelligent Data Science Lab starting Summer 2021. Currently, I am a Research Intern at the School of Computer Science at CMU, working on multiple projects.

profile photo

Oct 20 - Now:  CMU Research Intern  Continual Learning, 3D-Object Classification, Biologically-Inspired Computation, Agent Learning in Open-Endedness
July 21 - Jul 22:  MITACS Fellow, UBC Canada  Data-efficient Learning, Sparse representations, In-silico methods
Aug 20 - Jul 22:  Thapar Artificial Intelligence Lab  Multimodal Learning, Incremental Learning, Fuzzy-Decision making, Cognitively-inspired Computation
Apr 21 - Mar 22:  PathCheck (MIT Spin-off)  Time Series Forecasting, Statistics, Model-error prediction
Summer 21:  SURGE Fellow, IIT Kanpur  Design, Fabrication, Experimentation
Jan 21 - Jun 21:  Tata Motors  Transfer Learning, Neural Networks for the Shopfloor
Oct 20 - Mar 21:  NUS Research Intern  Self-Supervised Learning, Sim2Real Transfer, Domain Randomization
July 20:  Bharat Heavy Electricals Limited  Computer-aided design, Modeling, Optimization
Summer 20:  IIT Kanpur Research Intern  Deep Learning, Computer Vision, Natural Language Processing

Major Academic Achievements and Awards

  • Selected as a delegate at the 30th Harvard Project for Asian and International Relations (HPAIR)
  • Our work 'Accuracy of US CDC COVID-10 Forecasting Models' got featured in News-Medical.net
  • Our work on 'Deep-Precognitive Diagnosis' got accepted in IEEE Access.
  • Recipient of Globalink Graduate Fellowship that provides financial support to pursue full Master's/ PhD, or Postdoc at any MITACS partner University. (CAD$ 15000)
  • Awarded Dean's Merit List Scholarship for 2 consecutive years (AY-19/20 & 20/21) worth 50% Tuition Fee for being in Top 03% dept. ranks (US$ 4000).
  • Received MITACS Globalink Research Award to pursue project at Univ. of British Columbia (CAD$ 1200).
  • Selected to deliver a 'Lightening Talk' at NeurIPS 2021: ML for Public Health.
  • Awarded Students Undergraduate Research Graduate Excellence 2021 by IIT Kanpur (<4% acceptance).
  • Winner of the University of Queensland Engineering Design Hackathon, 2020.
  • Best Research Paper Award at the ICCAME, 2021 Conference.
  • Achieved All India Rank 1 in ICSE 2015 Computer Science exam out of 0.16 Million applicants.
  • Secured World Rank 98 in the International Olympiad of English Language 2016, Special achievement award.
  • Secured World Rank 264 in the International Olympiad of Mathematics 2015, Special achievement award.
  • Cleared the Uttar Pradesh State level of National Talent Search Examination (NTSE) 2015.
  • Bagged two Bronze Medals in 6th Intl. Young Mathematicians' Convention in both individual and team contests.

  • Under Review/ Upcoming Research Works

    Accuracy of US-CDC COVID-19 Forecasting Models [Preprint] [Press  PathCheck
    Aviral Chharia, G Jeevan, R A Jha, M Liu, J Berman, C Glorioso
    medRxiv/ Nature Scientific Reports (IF=4.380) [Under Review]

    Currently there is no universally recognized metric for evaluating performance of COVID-19 case-forecasting models. Such an evaluation is challenging due to variations in time for which different models were active, difference in forecast periods, model type, etc. How can we evaluate these effectively?

    Schizo-Net: A novel Schizophrenia Diagnosis framework using late fusion multimodal deep learning on Electroencephalogram-based Brain connectivity indices   TAiL
    N Grover, Aviral Chharia, R Upadhyay, L Longo
    IEEE Transactions on Neural Systems & Rehabilitation Engg. (IF=3.802) [Under Review]

    Schizophrenia is a serious mental condition that causes hallucinations, delusions, and disordered thinking. We present Schizo-Net, a novel schizophrenia diagnosis model based on multimodal late fusion of brain connectivity indices estimated from patient's EEG-activity. The study identifies Schizophrenia-related changes that occur in connectivity, and draws their vital significance to identify relevant biomarkers. Schizo-Net surpasses all current models and achieved state-of-the-art performance.

    EEG-based Non-invasive P300 Brain Computer Interface with Visual Stimuli ERP: A PRISMA compliant systematic review   TAiL
    J Kalra, P Mittal, N Mittal, A Arora, U Tewari, Aviral Chharia, R Upadhyay, V Kumar, L Longo
    IEEE Transactions on Neural Systems & Rehabilitation Engg. (IF=3.802) [In Prep.]

    This review article is aimed to comprehend the research trends in area of P300-based BCI design and provide direction for future developments. The study reviews current and future trends and discusses how these trends may impact researchers and practitioners alike in the future. The articles were classified based on a scheme consisting of research orientation and domains. Our results show considerable growth in the research interest in the field of BCI spellers using P300.

    ADHD-Net: EEG-aided Frontal Cortical loss detection for ADHD diagnosis in children on Continual mental task   TAiL
    Aviral Chharia, R Upadhyay, V Kumar
    IEEE Sensors Journal (IF=3.301) [In Prep.]

    ADHD is one of the most common heterogeneous neuro-developmental disorders of childhood. Children with ADHD have trouble paying attention, controlling impulsive behaviors, or being overly active. Here, we designed a Convolutional Time-Frequency domain Neural Network for Frontal Cortical loss detection on EEG-signals for high accuracy automated ADHD diagnosis. ADHD-Net surpases all current models and achieved state-of-the-art performance.

    Accepted Research Papers/ Preprints

    Deep-Precognitive Diagnosis: Preventing Future Pandemics by Novel Disease Detection with Biologically-inspired Conv-Fuzzy Network [PaperCMU
    Aviral Chharia, R Upadhyay, V Kumar, C Cheng, J Zhang, T Wang, M Xu
    IEEE Access (IF=3.367) #1 Journal in Engg. & Computer Science (General) on Google Scholar

    Currently there exists no Computer-Aided Diagnosis model that can itself Discover a novel disease, alert the radiologist and update its feature space on subsequent test samples in real-time. Can we address this novel task?

    Challenges in Equitable COVID-19 Vaccine Distribution: A Roadmap for Digital Technology Solutions [Paper  PathCheck
    J Bae, D Gandhi, J Kothari, S Shankar, J Bae, P Patwa, R Sukumaran, Aviral Chharia, S Adhikesaven, S Rathod, I Nandutu, Sethuraman TV, V Yu, K Misra, S Murali, A Saxena, K Jakimowicz, V Sharma, R Iyer, A Mehra, A Radunsky, P Katiyar, A James, J Dalal, S Anand, S Advani, J Dhaliwal, R Raskar
    arXiv Preprint

    In this early article, we identify challenges in logistics, health outcomes, user-centric matters, and communication associated with disease-related, individual, societal, economic, and privacy consequences. Primary challenges include difficulty in equitable distribution, vaccine efficacy, duration of immunity, multi-dose adherence, and privacy-focused record keeping to be HIPAA compliant. While many of these challenges have been previously identified and addressed, some have not been acknowledged from a comprehensive view accounting for unprecedented interactions between challenges and specific populations. Given these complicated issues, the importance of privacy-focused, user-centric systems for vaccine education and incentivization along with clear communication from governments, organizations, and academic institutions is imperative.

    A novel fuzzy approach towards in silico B-cell epitope identification inducing antigen-specific immune response for Vaccine Design [Paper] [Slides] [Video]  
    MITACS Globalink Fellow  UBC
    Aviral Chharia, A Narayan
    21st IEEE International Conference on Bioinformatics & Bioengineering (BIBE) 2021

    Despite progress in deep learning, in silico B-cell epitope prediction has low levels of accuracy compared to NMR spectroscopy and X-ray structural analysis. Can we improve precision rates without larger datasets?

    Multimodal Cognitively-inspired Incremental Learning in Fused Feature Spaces for Prognosticating Breast Cancer [Paper] [Video] [Preprint]
    Oral Presentation
    Aviral Chharia, N Kumar
    Medical Image Computing & Computer Assisted Intervention (MICCAI) PRIME 2021

    Proposed a Multimodal Cognitively-inspired model that attained state-of-the-art accuracy, framing the task as an Incremental Learning Problem. Proposed approach allows the model to continually update its learned feature space on non-stationary multimodal data stream. Demonstrated the model's ability to learn complex relationships between different multimodal attributes, training on severly imbalanced and limited data by mapping it to a high-dimensional fused feature space.

    Novel fuzzy approach to Antimicrobial Peptide Activity Prediction: A tale of limited and imbalanced data that models won't hear [Paper] [Poster]  TAiL
    Aviral Chharia, R Upadhyay, V Kumar
    NeurIPS, 2021: AI for Science: Mind the Gaps Workshop

    Proposed a fuzzy intelligence model for short (<30 aa) AMP activity prediction, and its ability to learn on limited and severely skewed high-dimensional space mapping is demonstrated over a set of experiments. The proposed model significantly outperforms state-of-the-art ML models trained on the same data.

    Learning after Depolyment: The Missed tale of Supervision  [Video]
    Aviral Chharia, N Kumar
    NeurIPS, 2021: ML for Public Health Workshop Lightening Talk

    Proposed a new paradigm to address the challenges faced in designing current Computer-Aided Diagnosis Models

    From Convolutions towards Spikes: The Environmental Metric that the Community currently Misses [Paper] [Poster]   TAiL
    Aviral Chharia*, S Chauhan*, R Upadhyay, V Kumar
    NeurIPS, 2021: AI for Science: Mind the Gaps and Human-Centered AI Workshop

    The AI community being obsessed with state-of-the-art scores (80% papers NeurIPS) due to which the environmental metric of models remains unreported. Here we proposed a novel evaluation metric, i.e., NATURE, for measuring the environmental performance of AI Models. Presented an in-depth attention highlighting the current gaps in Neuromorphic Computing and demonstrated mathematically why SNNs are better compared to traditional ANNs.

    Deep Recurrent Architecture based Scene Description Generator for Visually Impaired [Paper] [Video] [Slides]  TAiL
    Aviral Chharia, R Upadhyay
    12th IEEE Intl. Congress on Ultramodern Telecommunication & Control Systems 2020

    Developed a real-time assistance system to aid the visually impaired through recitation of surrounding scene descriptions generated using a Deep Recurrent Architecture combining CNN based feature extraction with Long-Short Term Memory (LSTM)

    Recent Trends in Artificial Intelligence-inspired Electronic Thermal Management [Accepted; In Press] [Preprint] [Video] [Certificate]
    Aviral Chharia*, N Mehta*, S Gupta*, S Prajapati*
    International Conference on Fluid Flow & Thermal Sciences (ICAFFTS) 2021

    Electronic Thermal management is a prerequisite for enhancing efficiency, lifespan and the prevention of overheating in electronic systems. This study presents an in-depth view of the recent models developed as an alterantive to traditional numerical approaches.

    A novel hybrid Fuzzy AHP-TOPSIS Approach towards Enhanced multi-criteria Feature-based EV Recommender System [Paper] [Slides]
    S Prajapati, Y Upadhyay, Aviral Chharia, B Sharma
    IEEE Global Conference on Advancement in Technology 2021

    This study presents a novel hybrid Fuzzy AHP-TOPSIS approach for muti-criteria feature-based recommender system.

    NeT-Vent: Low-Cost, Rapidly Scalable & IoT-enabled Invasive Mechanical Ventilator with adaptive control to reduce Pulmonary Barotrauma in SARS-CoV-2 patients [Paper] [Video] [Project WebsiteWinner-UQ Hackathon
    Aviral Chharia*, S Chauhan*, S Basak*, B Sharma
    IEEE Global Conference on Advancement in Technology 2021

    Pulmonary Barotrauma is a major cause of Ventilator-Induced Lung Injury in COVID-19 patients. Can mechanical design and control system improvements address this challenge in a cost-constrained scenario?

    Computational fluid dynamics-based disease transmission modeling of SARS-CoV-2 Intensive Care Unit [PaperBest Paper Award
    S Prajapati, N Mehta, Aviral Chharia, Y Upadhyay
    International Conference on Contemporary Advances in Mechanical Engineering 2021

    Analysed the SARS-CoV-2 disease transmission in an Intensive Care Unit and highlighted the flow of aerosol particles considering the combined as well as individual HVAC effects. Obtained results emphasized that aerosol particle flow has a promising application in sanitizing ICUs.

    Computational modeling and conjugate heat transfer study for in situ design of artificial porous media at varied orientations [Paper]
    S Prajapati, Aviral Chharia, N Mehta, S Yadav
    International Conference on Contemporary Advances in Mechanical Engineering 2021

    Investigated the characteristics of conjugate heat transfer on various orientations of computationally modeled artificial porous media structure useful in biomedical applications. The study establishes that for both horizontal and vertical orientations, an increase in porosity results in an increase of heat transfer rate from the porous media.

    Industrial Experience

    Industry4.0, Neural Networks for the Shop Floor [Certificate] [Slides]
    Tata Motors Limited
    Guide: Er. V Tyagi, Prof. A Sharma

    Lead the project that implemented Industry 4.0 integration of Coolant Dispensing Machine on Assembly Line which led to a cost saving of 0.165 million USD/year. Installed & analysed DURR wheel alignment machine for high-precision, contactless vehicle geometry determination & its subsequent Correction at Axle Level. Developed an automatic visual inspection system for quality control by identifying & classifying surface defects in steel strips using CNNs. Performed pedictive maintenance for downtime reduction on Assembly Lines.

    Modeling, Analysis & Optimization of Engine Components by BMX Module. CNC Machine Tool selection & application in machining IC Engine Components
    Bharat Heavy Electricals Limited
    Guide: Er. N Singla

    Implemented design methodology and modeled the crankshaft and piston of IC-Engine in PTC-Creo. Performed sensitivity analysis on crankshaft to determine critical dimensions affecting its mass and CG and find minimum balancing weight. Performed literature review for selecting tools based on tool-life and other factors along with developing GM code for machining these parts on CNC Machine.

    Other Research/ Technical Projects

    Brain Tumor Detection on MRI scans with Transfer Learning [Code]
    Self Project

    Implemented Transfer Learning on VGG-16 Convolutional Neural Network (CNN) with fine-tuning to detect whether a subject has Brain Tumor using Magnetic Resonance Imaging (MRI) scans, for assistance in Robotic Surgery. Performed data augmenation to increase training size data making use of limited data and hyperparameter tuning for decreasing model loss.

    Automated COVID-19 diagnosis on CXR scans using Convolutional Neural Network with end-to-end training [Code]
    Self Project

    Proposed a CNN based model for the automated detection of COVID-19 over CXR scans to aid radiologists. Trained the CNN from scratch on COVID-19 Radiography dataset, performed hyperparameter tuning, and achieved state-of-the-art 98% accuracy on the validation set with respective to Radiologist's clinical findings.

    Residual Attention Learning based Self-Supervised Network for Sim2Real Shape-Tracking and Pose Estimation of 6-DOF Origami-inspired Endoscopic Worm Robot
    National University of Singapore
    Code to be out soon

    Origami-inspired robots have drawn immense attention in recent years due to their shape morphing ability that is useful in a wide range of medical applications. However, shape-invariant pose estimation remains a challenge and is vital to study, control and automate the locomotion of origami robots. Proposed a Residual attention learning-based network trained using a self-supervised approach on domain randomized Endoscopic synthetic data to infer the robot’s pose in real-world data.

    Autonomous Vehicle based on Dublin's Luas Light Rail System [Code]
    Guide: Prof. Karamjit Singh & Prof. Amit Mishra, Course: UTA014

    Developed an autonomous vehicle using Arduino (C++ Programming) and remote wireless supervisory control (XCTU and XBee) with capability of ultrasonic obstacle detection and avoidance, self-parking, stopping at gantries in its path and safely co-existing with other vehicles. Designed IR module, Transmitter and Receiver circuits and fabricated it on PCB.

    Behavioral Cloning: Implementing NVIDIA's Self-Driving Car [Code]
    Thapar Developer's Student Club

    Developed an end-to-end Self-driving car using CNN to map pixels from front-camera to steering angles on a simulator. This deep learning approach required minimum training data & the system learned to steer, with or withoutlane markings, on both local roads and highways, even with unclear visual guidance in various weather conditions.The vehicle could identify traffic signs and avoid collisions.

    Maximum Equivalent & Principle Stress Reduction on Aircraft Wing using Carbon and glass fibre-reinforced composites with varying Spar Cross-sections
    Guide: Prof. N Grover

    Performed design, modeling and structural analysis of Beams, Cylinders, Plates and shells. Designed a cost-effective and lightweight aircraft wing (NACA- 4412), using carbon- and glass- fibre-reinforced plastic (CFRP and GFRP) composites. Modeled wing internal structure in SOLIDWORKS and performed the structural and vibration analysis in ANSYS with comparative study of spars of different cross-sections.

    Featured Talks

    Technical Skills

    • Programming Languages: C, C++, JAVA, Python, MATLAB, HTML
    • Libraries and Research Tools: Scikit-Learn, Pandas, PyCaret, NumPy, Seaborn, Matplotlib, Git, LATEX, Inkscape
    • Softwares and Simulations: OpenCV, Tensorflow, Keras, SOLIDWORKS, PTC-Creo, AutoCAD, ANSYS

    Jan 21-now:  Senior Student Mentor at Thapar Developer Student Club   Led a team of 5 students to boost the coding culture on campus. Mentored 12 undergraduates on Computer Vision projects. Conducted Workshops on Mechatronics systems, Localisation, Path Planning, Image Processing, and ML. Part of team that organized Hackathons on campus
    Jun 20-Dec 20:  Volunteer at Thapar Developer Student Club  Behavioral Cloning: Implementing NVIDIA's Self-driving car
    Apr 15-Apr 17:  Student Mentor at NavSrijan  Designed and taught a basic Computer and English speaking course to first time school students of the under-privileged section in society
    Apr 15-Apr 17:  Student Volunteer at Design for Change  Surveyed a local migrant colony from Bihar, interviewing more than 40 families to understand the barriers faced by these communities in education
    Apr 13-Apr 15:  Volunteer at Hope Initiative Society  Stressed on health awareness and the importance of cleanliness in neighbourhood

    This country counter seemed fun. It shows Page visit Counts since September 19, 2021.

    Template from Jon Barron. Thanks for Visiting :)
    Last updated: June 15, 2022