πŸ“š Projects | Research

Tags:  

Smart Home using HTTP/MQTT protocols

Designed HTTP based RESTful APIs for establishing end-to-end communication with lightweight digital signature security mechanism for Smart Home implementation

Workflow APIs for end to end 5G testbed

Designed Client Apps, Device App, UALCM, MEC Orchestrator, MEC Platform Manager and deployed Flask based RESTful APIs for intercommunication for running MEC apps on the 5G testbed .

CNN LSTM Attention based ECG Signal Classification

Employed CNN+Bidirectional LSTM+ Attention Model for classifying varied length 12 Lead ECG signals (Dataset: PhysioNet/CinC 2020) into 9 different types of cardiac arrhythmia with an f1-score ~ 0.8. CNN was used for a feature extraction tool in the normalized input signal. LSTM was later used with attention to identify patterns in the ECG data.

Ensemble of Deep Learning Architectures for Driver Drowsiness Detection

The proposed ensemble architecture consists of two deep CNN models of InceptionV3 that exclusively perform eye and mouth subsample feature extraction in the face picture. The two networks' output is transferred using stacking to the substructure logic of the ensemble. The device performance decides the state of the driver, along with a probability ranking, viz., drowsy and non-drowsy.

Nuggets for NLP - Predicting the most probable next word for a given word

Given a corpse C, the objective is to develop an algorithm that predicts the most expected next word W' to a given word W based on frequency analysis of W & W' in C. The algorithm finds by the basyesian probability P(W'|W) defined by the function f(W,W') that finds the probabilty of the occurence of W' given an immediate priori occurence of W in C. The algorithm is extended to learn on the transition matrix so generated to find n sentences (with or without semantics) given a seed word W.

Corpse Parser and Tokenizer for random tweets

The algorithm parses a corpse of tweets separated in lines to detect hashtags, questions, excalmation marks, urls, numbers (floating point) and negation words. the tokenizer then classifies each tweet under individual categories.

Digital Signature Scheme (RSA+MD5)

The C++ Implementation of my Digital Signature Algorithm featuring Encryption/Decryption using RSA Algorithm and Hashing Scheme using MD5 Digest. Let's keep the in-built cryptographic libraries at bay.

⏰ Journey (So far!)

πŸ“ƒ Publications

@inproceedings
{9352907, author={Kherani, Arzad Alam and Shukla, Gaurav and Sanadhya, Shashvat and Vasudev, Neha and Ahmed, Muneeb and Patel, Ashish Singh and Mehrotra, Rashi and Lall, Brejesh and Saran, Huzur and Vutukuru, Mythili and Singh, Abhishek and Seshasayee, Sushila and Viswakumar, Vinodh R and Loganathan, Kishore}, booktitle={2021 International Conference on COMmunication Systems NETworkS (COMSNETS)}, title={Development of MEC system for indigenous 5G Test-Bed}, year={2021}, volume={}, number={}, pages={131-133}, doi={10.1109/COMSNETS51098.2021.9352907}}

πŸ“· Fresh food on my hobby plate

Drop me a line πŸ“§

On most (almost all) of the Saturdays and Sundays I am available for discussion about ideas in applied deep-learning and collaborative robotics (over video). If I am otherwise free, I can spend some time over coffee to seek and learn higher order understanding of the purpose of life. In all the cases, my preferred primary contact is email. If you want to talk to me about these, my email address will be somewhere in my bio (A copy-paste of my email address might not guarantee success. You will need to type it). I read almost all of my emails yet sometimes I may not respond immediately. Pardon me for that but I will try to get back to you as soon as it'd be possible for me. After we familiarize I'd be happy to share my Telegram/Signal Id.

Copyleft. πŸ’» 2021. Muneeb :)