Hi, I am SushmitI'm a machine learning enthusiast/advocate currently appointed as a Research Engineer at Maya, an AI assisted healthcare platform. Previously I worked as a Research Engineer at the mHealth Research Group, Department of Biomedical Engineering, Bangladesh University of Engineering and Technology (BUET). I have completed my Bachelor's in Electrical and Electronic Engineering from BUET. My research interests include Computer Vision, Biomedical Signal Processing, Natural Language Processing, Domain Adaptation and Human Activity Recognition. I have been serving as a co-ordinator in Bengali.AI since 2018. We are creating standardized datasets to democratize the use of Bangla. In my free time I love reading poems and playing guitar. Have been trying to teach myself to play Accordion lately.
Runners Up @ Call4Nation Competition arranged by ICT Division, Peoples Republic of Bangladesh
Call4Nation was an initiative by the government to accelerate start ups to tackle many challenges during the pandemic. Our ‘RadAssist’ teleradiology platform won the Runner’s Up prize under the Health Care Equipment & Treatment category. RadAssist is the first AI-based teleradiology solution developed in Bangladesh and is currently operational having collaboration with multiple hospitals dealing with the pandemic.
Public Lecture on 'Demystifying Artificial Artificial Intelligence'
Got invited to Dhaka University Trust Reading Club to deliver a lecture on demystifying artificial intelligence. Students and enthusiasts from different disciplines attended the talk. News Link
Hosted First Kaggle Featured Competition from Bangladesh
A host of of the first Kaggle featured competition from Bangladesh that was held on a dataset co-designed by me. Raised 50k USD in funds for the competition. Got an additional 120k USD in waiver. Competition was launched on 19th December, 2019. 2059 teams from all over the world participated in this competition with 39,058 submissions.
My research involves interdisciplinary projects incorporating deep learning. A key problem for biomedical engineering tasks is the scarcity and domain dependency of data. Achieving domain agnostic feature extraction is a key to ubiquitous system. My research at MhealthLab, BUET is currently focused on constructing fine grain disease diagnosis based assistive systems that perform utilizes domain invariant features for biomedical image data.
Fine Grain Classification of Xray Images
Research on novel approach based fine grain classification of thoracic radiographic image classification network.
Human Activity Recognition from First Person Chest Mounted Camera Data
Construction of novel segmentation based two stream network for human activity recognition from first person video. The network uses segmentation masks of video frames as a replacement of flow based streams to accurately extract relevant features for this. Multiple attention based gates ensure the extraction of only the most relevant features. This work is a continuation of the research done by team BUET Synapticans on VIP Cup 2019 (ICIP, Taiwan 2019). I tutored the competition team of mHealth research lab and placed 3rd in the finals.
Sleep Stage Classification from Single Channel EEG
“End-to-end Sleep Staging with Raw Single Channel EEG using Deep Residual ConvNets”, BHI 2019
Biomedical Image Compression using Deep Learning
RNN based compression of X ray data. Undergraduate thesis. Published in BHI 2019
Lung Tumor Segmentation
Lung tumor segmentation from NCIS Radiomics Dataset using dilated convolution based scheme.
S. Alam, T. Reasat, A. Sushmit, S. M. Siddiquee, F. Rahman, M. Hasan, A. I. Humayun “Multi-label Classification of Common Bengali Handwritten Graphemes: Dataset and Challenge”, Modified version submitted to CVPR 2021, (A Large Multi-Target Dataset of Common Bengali Handwritten Graphemes), 2020
A. Sushmit, Partho Ghosh, Md.Abrar Istiak, Nayeeb Rashid, Ahsan Habib Akash, Taufiq Hasan, “SegCodeNet: Color-Coded Segmentation Masks for Activity Detection from Wearable Cameras.”, Submitted to IEEE Transactions on Image Processing, 2020
P. Ghosh, A. Istiak, N. Rashid, A. H. Akash, R. Abrar, A. G. Dastider, A. Sushmit, T. Hasan, “Privacy-Aware Activity Classification from First Person Office Videos”, Arxiv Preprint, June, 2020
A. Sushmit, S. Zaman, AI Humayun, T. Hasan, MIH Bhuiyan, “X-Ray Image Compression Using Convolutional Recurrent Neural Networks”, IEEE BHI, Chicago, May, 2019
S. Hossain, S. Najeeb, A. Sushmit, Z. Rahabin, M. Ariful Haque, “A Pipeline for Lung Tumor Detection and Segmentation from CT Scans Using Dilated Convolutional Neural Networks”, ICAASP, Brighton, May, 2019
AI Humayun, A. Sushmit, T. Hasan, MIH Bhuiyan, “End-to-end Sleep Staging with Raw Single Channel EEG using Deep Residual ConvNets”, IEEE BHI, Chicago, May, 2019
Awards And Honors
Runners Up @ Call4Nation2020
Call4Nation Competition (Pandemic Crisis Management) arranged by ICT Division, Peoples Republic of Bangladesh
2nd Runners Up, IEEE VIP Cup 20192019
IEEE Video Image Processing Cup, ICIP 2019. https://signalprocessingsociety.org/get-involved/video-image-processing-cup
Champion, IEEE VIP Cup 20182018
Lung tumor detection and segmentation from 3D lung CT scans from the NSCLC-Radiomics Dataset
Travel Grant, ICIP 20182018
Received IEEE SPS Travel Grant to attend ICIP 2018 conference in Greece
GPU Grant, NVIDIA2018
Obtained a Titan Xp GPU as a research grant from NVIDIA under their GPU Grant program for data science researchers
Champion, National Robotics Competition, 20172017
Designed a Deep Residual Architecture based network to successfully perform automated Bangla Handwritten Digit Recognition
Champion, National Power Energy Hackathon2017
Represented the team ‘BUET Luminaries’ and designed an automated national grid load prediction system using machine learning for successful integration of local smart grid systems with SCADA network.
2nd, Divisional Round, BdPhO 20122013
Bangladesh Physics Olympiad: Divisional Round. BdPhO 2012. Attended extended camp for the preparation of international round.
2nd, National Round, BDSO 20122012
Final Round of Bangladesh Academy of Science Olympiad 2012, BUET.