Ph.D. @ Data Science Institute, NUIG | Researcher @ Newcastle University & Confirm Smart Manufacturing | Ex ML Infrastructure Engg @ ARM | Ex Researcher @ AI Institute, UofSC | Ex Embedded System Engg @ Four Corners Tech.
Design algorithms to improve the Resilience, Interoperability, Scalability (RIS) of IoT devices.
Deep optimization, deployment, and efficient execution of a wide range of ML models on AIoT boards, small CPUs, and MCUs based devices.
Design resource-friendly algorithms that enable tiny devices to locally re-train themselves on-the-fly (after deployment) using the unseen real-world data patterns.
Design compression algorithms to improve communication efficiency when distributed global ML model training and when collaborative learning in IoT.
In 2020 - 2021 (1 year), I contributed to science by publishing 15+ first-author full papers in top-tier venues such as IEEE Internet Computing, IEEE IoT Journal, ECML PKDD, ACM IoT, IEEE SCC, IEEE UIC, IEEE BigData. Also provided 10+ demos, short papers at IEEE PerCom, ACM/IEEE ICCPS, ACM/IEEE IoTDI, IEEE WF-IoT, ACM SenSys, ACM Middleware, AAAI, ACM UbiComp-ISWC. I obtained my Masters from NUI Galway in Electronics and Computer Engineering. My Master's project was supervised by Prof. Peter Corcoran.
[Nov-2021] EPE Champion award - communicated scientific breakthroughs to over 2000+ audiences
[Apr-2021] ECML PKDD 2021, our 'Machine Learning Meets Internet of Things: From Theory to Practice' tutorial accepted [Link][Tutorial Website]
[Mar-2021] Security, Privacy and Trust in the Internet of Things (SPT-IoT) workshop at PerCom '21. Presented our 'Edge2Guard: Botnet Attacks Detecting Offline Models for Resource-Constrained IoT Devices' paper [Link]
[Mar-2021] PerCom 2021, presented our paper 'Ultra-fast Machine Learning Classifier Execution on IoT Devices without SRAM Consumption' [Link]
[Feb-2021] Our research featured on the Confirm Smart Manufacturing website [Link]
[Jan-2021] Our COVID-away paper is also made available at the World Health Organization's global literature on coronavirus disease page [Link]
[Oct-2020] Presented papers "RCE-NN and Edge2Train" in IoT Conference [Pics][Pics]
[Oct-2020] Our "Avoid Touching Your Face" paper won the Second place in IoT-HSA workshop [Link]
[Aug-2020] Presented "Adaptive Strategy to Improve the Quality of Communication for IoT Edge Devices" in the institute meeting at the Data Science Institute, NUI Galway.
[Aug-2020] Attended the IEEE Virtual World Forum on Internet of Things 2020 - Tutorial Program [Cert]
[Jan-2020] First flight of my first DIY drone. Features: Altitude hold, auto level, GPS lock, and return to home [Video]
[Dec-2019] Presented poster "Smart speaker design and implementation with biometric authentication and advanced voice interaction capability" at 27th AIAI Irish Conference on Artificial Intelligence and Cognitive Science [Poster][Pics]
[Dec-2019] Presented paper "AI Vision: Smart speaker design and implementation with object detection custom skill and advanced voice interaction capability" at 11th IEEE - International Conference on Advanced Computing (ICoAC)
[Apr 2019] Provided a bench demo of MEngg project at NUIG
[Mar 2019] Presented poster for our "Assess, Respond, Monitor, Strengthen Glove (ARMS glove) for stroke" at Blackstone Launchpad, NUIG
[Feb 2019] Presented MEngg project poster titled "Design and realization of a wireless smart-speaker" at NUIG [Poster]
According to CORE Rankings Portal:
[A*] - flagship conference, a leading venue in a discipline area. [A] - excellent conference, and highly respected in a discipline area.
[B] - good to very good conference, and well regarded in a discipline area. [C] - other ranked conference venues that meet minimum standards.
2022: First author. [B] OTA-TinyML: Over the Air Deployment of TinyML Models and Execution on IoT Devices @ IEEE Internet Computing Journal (IEEE IC) [Repo] [Paper yet to appear].
2022: First author. [B] ElastiQuant: Elastic Quantization Strategy for Communication Efficient Distributed Machine Learning in IoT @ The 37th ACM/SIGAPP Symposium on Applied Computing (ACM SAC) [Paper yet to appear].
2021: First author. Imbal-OL: Online Machine Learning from Imbalanced Data Streams in Real-world IoT @ IEEE International Conference on Big Data (IEEE BigData) [Paper].
2021: First author. [A] Enabling Machine Learning on the Edge using SRAM Conserving Efficient Neural Networks Execution Approach @ European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD) [Paper].
2021: First author. [9.5] Impact Factor. ML-MCU: A Framework to Train ML Classifiers on MCU-based IoT Edge Devices @ IEEE Internet of Things Journal (IEEE IoTJ) [Paper] [Repo].
2021: First author. [B] Train++: An Incremental ML Model Training Algorithm to Create Self-Learning IoT Devices @ International Conference on Ubiquitous Intelligence and Computing (IEEE UIC) [Repo] [Paper].
2021: First author. [B] Globe2Train: A Framework for Distributed ML Model Training using IoT Devices Across the Globe @ International Conference on Ubiquitous Intelligence and Computing (IEEE UIC) [Paper].
2021: First author. [A] An SRAM Optimized Approach for Constant Memory Consumption and Ultra-fast Execution of ML Classifiers on TinyML Hardware @ IEEE International Conference on Services Computing (IEEE SCC) [Paper].
2021: First author. [B] Towards Distributed, Global, Deep Learning using IoT Devices @ IEEE Internet Computing Journal (IEEE IC) [Paper].
Demo, Short Paper, and Tutorial
2022: [A*] TinyFedTL: Federated Transfer Learning on Ubiquitous Tiny IoT Devices @ The 20th International Conference on Pervasive Computing and Communications (PerCom) [Repo] [Paper yet to appear].
2022: First author. [A*] Training up to 50 Class ML Models on 3$ IoT Hardware via Optimizing One-vs-One Algorithm @ Association for the Advancement of Artificial Intelligence (AAAI) [Repo] [Paper yet to appear].
2021: First author. [A] Machine Learning Meets Internet of Things: From Theory to Practice @ European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD) [Paper] [Tutorial Website].
2021: First author. [A*] ElastiCL: Elastic Quantization for Communication Efficient Collaborative Learning in IoT @ The ACM Conference on Embedded Networked Sensor Systems (ACM SenSys) [Paper].
2021: [A] GNOSIS- Query-Driven Multimodal Event Processing for Unstructured Data Streams @ The International Middleware Conference (ACM Middleware) [Paper].
2021: [A*] Air Quality Sensor Network Data Acquisition, Cleaning, Visualization, and Analytics: A Real-world IoT Use Case @ The ACM international joint conference on pervasive and ubiquitous computing (ACM UbiComp-ISWC) [Paper] [Repo].
2021: First author. Ensemble Methods for Collaborative Intelligence: Combining Ubiquitous ML Models in IoT @ IEEE International Conference on Big Data (IEEE BigData) [Repo].
2021: First author. TinyML Benchmark: Executing Fully Connected Neural Networks on Commodity Microcontrollers @ IEEE World Forum on Internet of Things (IEEE WF-IoT) [Paper] [Repo].
2021: First author. Porting and Execution of Anomalies Detection Models on Embedded Systems in IoT @ ACM/IEEE Conference on Internet of Things Design and Implementation (ACM/IEEE IoTDI) [Paper].
2021: First author. SRAM Optimized Porting and Execution of Machine Learning Classifiers on MCU-based IoT Devices @ International Conference on Cyber-Physical Systems (ACM/IEEE ICCPS) [Paper].
2021: First author. Edge2Guard: Botnet Attacks Detecting Offline Models for Resource-Constrained IoT Devices @ PerCom Workshop (SPT-IoT) [Paper] [Repo].
2021: First author. OWSNet: Towards Real-time Offensive Words Spotting Network for Consumer IoT Devices @ IEEE World Forum on Internet of Things (IEEE WF-IoT) [Paper].
2020: First author. Avoid Touching Your Face: A Hand-to-face 3D Motion Dataset (COVID-away) and Trained Models for Smartwatches @ Internet of Things based Health Services and Applications (IoT-HSA) [Paper] [Repo].
2020: First author. RCE-NN: A Five-stages Pipeline to Execute Neural Networks (CNNs) on Resource-constrained IoT Edge Devices @ International Conference on the Internet of Things (ACM IoT) [Paper].
2020: First author. Edge2Train: A Framework to Train Machine Learning Models (SVMs) on Resource Constrained IoT Edge Devices @ International Conference on the Internet of Things (ACM IoT) [Paper] [Repo].
2020: First author. Adaptive strategy to improve quality of communication for IoT edge devices @ IEEE World Forum on Internet of Things (IEEE WF-IoT) [Paper].
Slides of recent technical talks.
2021: Imbal-OL 15 min talk @ BigData. [Pdf Slides]
2021: Efficent-NN-Execution 20 min talk @ ECML PKDD Main Conference. [Pdf Slides]
2021: Edge2Guard 15 min talk @ PerCom SPT-IoT. [Pdf Slides]
2020: Covid-away-Wearables 20 min talk @ ACM IoT-HSA. [Pdf Slides]
Researcher and Co-Lead - Edge Computing, IoT
CONFIRM SFI Research Centre for Smart Manufacturing, Ireland. May 2019 - May 2022 (3 years)
In addition to cutting edge research work and my 25+ publications at premium journals and conferences, I created high research impact by bringing in and leading multiple collaborations (top 1% ranked scientists) from Industry (ARM, Ayyeka, EdgeImpulse, etc.), research organizations (CRT-ML, AI Institute UofSC, Lero Research Centre, etc.), and Accademia (Newcastle Uni, Cardiff Uni, TU Wien Uni, Swinburne Uni, etc.). Below are a few interesting research projects that I contributed to:
Built 'COVID-away models' to reduce the spread of the current global pandemic. When any of our models are deployed on smartwatches, it can trigger a timely notification (e.g. vibration) when the hand of the smartwatch user is moved (unintentionally) towards the face [Repo]. This work is featured on the Confirm website [Link] and made available on the WHO's global literature on coronavirus disease page [Link]. Also won Second place in the IoT-HSA'20 workshop [Link].
Designed 'Adaptive Strategy' to improve the quality of communication for IoT edge devices. When devices within an IoT system are equipped with our strategy, they can adapt according to dynamic context whilst ensuring the highest level of communication quality, thus, improving the overall resilience of the entire IoT system [Paper].
Provided demo at WF-IoT'21 conference to present audience the 'TinyML benchmark process by executing fully connected neural networks on commodity microcontrollers' [Paper].
Co-authored the 'GNOSIS' project and assisted in demonstrating 'Query-Driven Multimodal Event Processing for Unstructured Data Streams' at the ACM Middleware'21 conference [Video][Paper].
Presented an interactive poster titled 'ElastiCL' at SenSys'21 conference to show audience 'how to perform elastic quantization for communication efficient collaborative learning in IoT' [Paper].
Collaborated with researchers from Unimore University, Italy, and contribute to a real-world IoT use case, whose outcome was an interactive poster titled 'Air Quality Sensor Network Data Acquisition, Cleaning, Visualization, and Analytics' presented at UbiComp-ISWC ’21 [Paper][Poster].
Designed resource-friendly algorithms named Edge2Train [Paper][Code], Train++ [Paper][Code], ML-MCU [Paper][Code] that enable tiny devices to locally re-train themselves on-the-fly (after deployment) using the unseen real-world data patterns.
Explained to AAAI'22 conference audience the secret behind 'training up to 50 class ML models on 3$ IoT hardware' [Repo].
Other tasks with Confirm include developing, delivering Education and Public Engagement (EPE) activities; preparing quarterly project reports and for Confirm annual conference; presenting in networking sessions; attending meetings such as monthly R&D seminars, grant calls ERC, Horizon2020, Marie Skłodowska-Curie, Trustworthy AI, etc.
Researcher - Collaborative ML, Distributed ML
Newcastle University, Newcastle upon Tyne, England. Sept 2021 - May 2022 (8 months)
My main duties include the following: Design compression algorithms to improve communication efficiency when distributed global ML model training and when collaborative learning in IoT; Contribute to the preparation of grant proposals and the assessment of research findings for potential commercial exploitation; Dissemination of the conducted research results for relevant research stakeholders in the form of live demo and ppt slides; Supervise and provide monthly mentoring and technical assistance to research graduates, students and other junior members associated to the research group. Initial outcomes and work in progress are listed below:
Published paper titled 'Toward Distributed, Global, Deep Learning using IoT Devices' at IEEE Internet Computing Journal [Paper].
Paper in press 'ElastiQuant: Elastic Quantization Strategy for Communication Efficient Distributed Machine Learning in IoT' at The 37th Symposium on Applied Computing (ACM SAC).
Paper in press 'OTA-TinyML: Over the Air Deployment of TinyML Models and Execution on IoT Devices' at IEEE Internet Computing Journal [Repo].
Delivered a 40 minutes lecture to cover 'Osmotic Distributed IoT Training' [Pdf Slides].
Draft stage paper 'Multi-Component Optimization and Efficient Deployment of Neural-Networks on Resource-Constrained IoT Hardware' [Repo].
Machine Learning Infrastructure (DevOps) Intern
ARM, Galway, Ireland. May - Oct 2021 (6 months)
ARM teams are entering a new growth phase to develop innovative technologies and products for new markets. By joining the ML Infrastructure Galway team, I contributed to ARM by; (i) enabling the ML software teams to ensure they successfully adopt the latest and most comprehensive DevOps practices; (ii) performing Cloud setup and Board Farm Management to enable globally distributed ARM research, engineering teams to craft software that powers the next generation of ML + CV based mobile apps, portable devices, home automation, smart cities, self-driving cars. Following are the principles/systems/devices/technology that I learned and worked with during the internship.
Setting Gerrit mirrors for repositories. Gerrit Code Review (usage of git commands such as checkout, push, pull, checkout, config, status, add, commit). Gerrit connection settings, replication events.
Usage of HashiCorp Vault, Engpwdb (Engg Password Database). Create, store, use Jenkins API keys, tokens, secret text in Jobs.
Commissioning and setup of Single Board Computers (SBCs) such as Hikey960, Odroid, Raspberry Pi, Pumpkin i500. During hardware shipping - awareness of trade compliance, shipping regulations, and IP.
QEMU emulation of development boards inside both Windows and Ubuntu as hosts.
Create sites using Atlassian Confluence to document procedures and investigation findings of given tasks. Created the following sites: (i) Development boards emulation using QEMU. (ii) AWS EC2 Status by Project. (iii) Jenkins credentials creation, usage in groovy, handling .netrc files. (iv) Implementing Gerrit mirror. (v) Adding Hikey board as a CI agent.
Using ServiceNow for tasks such as resolve internal IT issues, obtain fixed IPs for dev boards, request details (hypervisor platform, FQDN, Rhev manager, etc.) of static hosts in use.
Automation using Ansible. After given hosts (fresh) setup in Nagios, 3lds, etc. adding it to AWX inventories. Then, running full setup jobs (from .yml code) for new hosts.
Flashing and rooting of Android smartphones to enable seamless Neural Network model performance benchmark on GPU in Pixel 4a. Usage of Magisk Manager, SDK Platform Tools, Team Win Recovery Project (TWRP), Odin, other tools.
Usage of Jenkins Pipeline (adding dev boards as agents), European Association for Language Testing and Assessment (EALATA), open source cloud computing infrastructure (OpenStack).
Scripting in Bash & Python, advanced debugging in Ubuntu environment. Exposed to using Groovy scripting.
Theoretical knowledge and basic usage of AWS CLI and services: Lambda, SNS, Load balancer, EC2, Optimization, Pricing, SQS, ECS, EKS, Fargate.
Setup and usage of Nagios (Industry-standard IT infrastructure monitoring) to monitor multiple parameters (load, temperature, uptime, kernel, ping, LDAP security check, etc.) on various types of dev boards, agents, servers, and local host machines. Tasks involve add hosts .cfg to Nagios server, add authorised key on client, install plugins on client (OS specific), set groupings, removing particular/specified Nagios checks/monitoring on the device, Slack-Nagios integration for alerts.
Daily usage of Atlassian JIRA for scrum sprints, create-assign-resolve-validate tickets, manage backlog tasks, issue tracking, prioritizing work, roadmap planning.
Contributed to preparing ARM teams for migrating jobs running on in-house machines to AWS (bakery setup to start using agents from AWS): Communication with global ARM teams, preparing roadmap, tracking status by team/project, understanding Groovy script of various teams, then alter by using sshagent in the groovy scripts instead of relying on private SSH keys on disk.
Usage of Git web interface, Git GUI, Git CLI, Git LFS, IntelliJ, Slack Workspace, Slack alerts, Slack Apps and Integrations - App directory.
Basic level implementation of Continuous Integration (CI), Continuous Delivery (CD), containerization technologies (Docker, Kubernetes), Assisted in CI/CD of other DevOps teams in ARM.
Worked in an Agile Team: Contributed to meetings such as stand-ups, retrospectives, refinements.
Researcher and Co-Lead - ML Solution Design
Artificial Intelligence Institute, University of South Carolina, Columbia, USA.Aug 2020 - Oct 2021 (1.2 years)
My responsibilities as a researcher were to drive impact by; (i) Designing use-case based ML Models, Deep Optimization, Hardware Deployment, and Efficient Execution; (ii) Collaborating with scientists and engineers from the cloud computing, semantic web, Industry 4.0 research areas; (iii) Supporting quick concept development of new and emerging ideas; (iv) Open sourcing high-quality code and reproducible results for the TinyML community and TensorFlow Lite, Micro users; (v) Publishing the performed state-of-the-art research work as papers in top quality conferences. Below is the list of projects with their outcomes that I contributed to.
Presented at ECML PKDD'21 conference 'an approach for efficient execution of already deeply compressed, large neural networks on tiny devices' [Poster][Paper].
Designed 'OWSNet' for consumer IoT devices, which is a real-time offensive words spotting network. OWSNet is designed to ensure a healthy verbal environment and avoid harmful incidents by policing the usage of language. [Paper].
Designed 'Edge2Guard', which are resource-friendly standalone botnet attacks detecting models that enable tiny devices to instantly detect IoT attacks without depending on networks or any external protection mechanisms [Repo].
Provided demos at popular conferences to present audience the process of 'porting and execution of ML models on tiny devices in IoT' [IoTDI'21 Demo][ICCPS'21 Demo].
Teaching Support Staff (TSS)
NUI Galway, Ireland. May 2019 - May 2022 (3 years)
I contributed to the below modules by teaching numerous technologies to students from various departments (CS, EEE, Industrial Engineering), year of study (first to final year), and programs (bachelors, masters, doctoral). My responsibilities include delivering lectures, organizing tutorial labs, creating assignments, providing feedback for student reports, and assisting lecturers.
2021: Lab supervisor, TSS for Tools and Techniques for Large Scale Data Analytics (CT5105) @ School of CS. Supervisor Dr. Matthias Nickles.
2021: Lab supervisor, TSS for Microprocessor Systems Engineering (EE224), Electrical Circuits and Systems (EE230) @ School of EEE. Supervisor Prof. Martin Glavin.
2020: Data visualization (CT5100), Web and Network Science (CT5113) @ School of CS. Supervisor Dr. Conor Hayes.
2019: Professional Skills - I (CT1112) and 2nd reader/examiner @ School of CS. Supervisor Prof. David O'Sullivan.
2019: Fundamentals of EEE - I (EE130), Fundamentals of Engineering (EI140) @ School of EEE. Supervisor Prof. John G Breslin.
R&D Embedded System Engineer
Four Corners Technologies (4CT) Pvt. Ltd, India.Oct 2016 - Nov 2018 (2.1 years)
At 4CT, we developed end-to-end IoT smart solutions for retails, workspace, kiosks, and outdoor billboards. I was the hardware guy in this firm, where my role was to design-build-program the wireless embedded system of IoT devices, then connect its data stream to cloud services hosting our in-house analytics engines. Below is the list of projects that I contributed to, which are currently deployed on a large scale and one of the high revenue sources for the firm.
Workspace Occupancy Monitoring: We designed a wireless embedded system with Panasonic Grid Eye thermal sensor to monitor the workspace occupancy. This occupancy data was sent to our web app to generate client requirement-based meaningful insights such as rich visualization of building and workspace utilization, detailed occupancy patterns, extensive reporting of occupancy by the department and by function, etc.
Remote Hoardings Monitoring: We designed an IP66 grade Linux-based IoT camera with 4G connectivity and integrated multiple outdoor LDR sensors. 250 of our IoT cameras were installed across the city and rural areas to monitor the outdoor billboards to provide view clarity, material quality, installation quality, pillar quality, lighting quality, live stream, etc. when requested by the billboard owners or clients via our billboard management system.
Retail Sense 'Progressive business decisions with live data at your Fingertips'. We designed Retail Sense, which is a low-cost camera-based wireless footfall people counter. The raw footfall count was sent to our web app, where it was converted into meaningful information that revealed patterns and profitable insight which is used to make key decisions on ideal staffing levels with placement based on the hour, day, month, season, facility’s layout and operations, etc.
e-Health Kiosks: We designed an MCU-based embedded system using sensors to accurately measure the height (using MaxBotix ultrasonic sensor), weight (using load cells mapped to a 24 bit ADC), and heart rate (using Max30100 pulse oximetry sensor). This board computes the height (cms) weight (Kgs) and heart rate (BPM & SPO2) and sends it to the system of the digital signage kiosk via USB.
R&D Intern - IoT Sensor Applications
Flamenco Tech India Pvt. Ltd.Apr - Sept 2016 (6 months)
Contributed to an end-to-end parking guidance system solution for a shopping complex client with a multi-story parking facility. My role was to configure and install hundreds of wireless ceiling mount ultrasonic sensors with its gateways for accurate parked car detection. Then I integrated the sensor data into the cloud-level analytics suite designed by other team members. After successful solution deployment, drivers could get real-time information about unoccupied parking spots and find one much quicker. The benefits for our client include reduced labor-intensive fixed message signs, reduced traffic congestion and pollution as we reduced search time for free spots, set dynamic hourly rate pricing based on real-time changes in demand and supply of parking spaces.
Embedded Hardware Software Co-design for ML-based Edge Computing
Drive embedded software (SW) development from initial concept to implementation, platform optimization, and performance validation. Skilled in designing and developing embedded SW for specialized in-house designed hardware/device. Worked closely with ML and hardware design teams to identify optimal SW architecture and implementation solutions.
Experienced using Digi’s wireless SOCs in networks; Intel Movidius NCS, Google Coral USB AI accelerator; SBCs including Raspberry Pi family of boards, Nvidia Jetson family, Intel NUC series, Google Coral, LattePanda boards; Arduino boards such as Pycom five network FiPy, STM32 blue pill, Espressif modules, Nordic SoCs; AIoT boards from Seeed studio; Feather and Wing series boards from Adafruit Industries.
Hands on prototyping experience using Panasonic’s PaPIRs, Grid-EYE infrared arrays; MaxBotix range finders; Maxim Integrated healthcare sensors; Thermoelectric Peltier modules; Interlink FSRs, Flex sensors; Melexis contactless IR temperature sensor; ST FlightSense ToF technology, ReSpeaker DSP mic-arrays, Leap Motion, variety of Grove sensors from Seeed studio.
Multi-sensor, wireless, low-power embedded system design using a range of ARM MCUs; using PICCCS, Keil, other embedded development tools, IDEs, debuggers (JTAG). Establish custom performance evaluation methodology, generate technical documentation and test procedures. Exposure to PCB layout, schematic capture, fab package release (Gerber, drill, BOM) to build mixed-signal hardware using Proteus or Eagle.
Experienced working in power-constrained typologies-based wireless environments; solid knowledge of wireless and wired communication systems; protocols and peripherals including BLE, Wi-Fi, LTE, GPS/GNSS, CoAP, MQTT, 6LoWPAN, Z-Wave, ZigBee, LoRaWAN, SigFox, AMQP, XMPP, HTTP/2.0, Digital, Analog, I2S, USB, UART, CAN, I2C, SPI, RS232, RS485.
Use-case based ML Model Design, Deep Optimization, Hardware Deployment, and Efficient Execution
A solid foundation on Neural Network components such as weights, activation function (such as Sigmoid, Tanh, Softmax, ReLU), different types of network layers (such as Conv1D, DepthwiseConv2D, MaxPool1D, Dropout, Average, Reshape), training, validation, testing process, gradients, partial derivatives, chain rule, backpropagation, optimizers (such as GD, SGD, mini-batch GD, Adam), loss functions. Experienced in implementing various types of NNs in Tensorflow Lite framework, and using libraries such as Numpy, SciPy, Pandas for precomputing.
Exposure to Deep Learning with classical computer vision techniques for image recognition, object detection plus tracking, acoustic scene identification. Firm knowledge in designing ML-based end-to-end edge analytics, signal processing, computer vision pipelines that are scalable, repeatable, and secure.
Experienced in optimizing model graph, size, workload, operations, performing quantization-aware training, and post-training quantization. Converting and stitching ML models with the main IoT application/program, followed by building executable binaries and deployment on any given hardware.
Design and training of use-case-based supervised and unsupervised ML algorithms for inference tasks such as forecasting, anomaly detection, classification. Then using frameworks such as emlearn, micromlgen, sklearn porter, m2cgen for code generation of the trained model and efficient execution on small CPU and MCU-based devices in IoT.
Solid understanding of popular models such as MobileNet, SqueezeNet, Inception V1, MnasNet, NASNet mobile, DenseNet, DeepLabv3, PoseNet, EAST, other TinyML models. Experienced in optimizing such models in multiple aspects, followed by their efficient deployment and execution on any target device.
Collaboration, Management for Cutting-edge Research
In my research career, I successfully initiated numerous collaborations with internationally top 1% ranked research scientists and engineers. The outcomes are 25+ papers (first-author in 20+) in prestigious venues. I communicated our collective scientific breakthroughs to over 2000+ technical audiences via invited talks, tutorials (at ECML PKDD'21), live demos (at IoTDI'21, ICCPS'21, WF-IoT'21, Middleware'21), interactive posters (at AAAI'22, SenSys'21, UbiComp-ISWC'21). In this process, I obtained the following skills.
Routinely work towards long-term ambitious research goals while identifying intermediate milestones and breaking down complex problems into approachable stages.
Literature review to investigate the advancements, then conduct research to advance state of the art and solve specific problems at scale in one or more of the following areas: Pervasive computing, Ubiquitous computing, IoT, Optimization Methods, TinyML, ML Systems, Edge Computing, Applied ML, Collaborative ML, Distributed ML.
Contribute research that can be applied to AI-IoT product development in a firm. Productionize and ship research into a firm's products, and thus to its users worldwide; knowledge and experience of data privacy by design best practice; researching and implementing novel ML and statistical approaches to add value to the business.
Ability to work independently, set up experiments, demonstrate progress through principled use of metrics. Define and conduct statistical hypothesis testing to ensure ML model reliability in production, plus to support experiments and inferences. The research I conduct follows the Reproducibility and Open Science principles to make my contributions (publications, data, code) and their dissemination accessible to all levels of society, amateur or professional, so they can repeat my proposed approach to reach similar conclusions.
Strong experience in presenting research project output to a wide range of audiences and demonstrating the ability to influence the audience through human-AI interactions and effective visuals. Communicating cross-functionally, identifying key partners, and building relationships with them. Placing a human-centered focus on the work (context, end-user impact), finding solutions that work in practice and have a significant impact.
Experienced leading globally distributed researchers and engineers in solving modeling problems using AI/ML approaches. Have managed research assistants, junior engineers, graduate research students by providing them technical guidance as needed. Managed communications including in-person and virtual meetings, to provide supervision, peer programming, training, and mentoring.
Participated in requirements sessions, architecture discussion, tech reviews, code reviews and contributed by providing inputs and steering the team/collaborators; Worked with product teams and stakeholders to identify problem statements that can be solved using data science, ML, and the automation of the process. Experienced in partnering closely with grant provides to build a holistic understanding of customers, products, and business.
Write Deployment Ready IoT Applications for Edge Devices in C, Embedded C, C++, Python
Firm experience in selecting cost and computational requirement-based hardware from a range of 8, 16, 32-bit MCUs, microprocessors, FPGAs. Then programming them to solve a wide range of problems in the given IoT use-case. Work closely with multiple teams to remotely (over-the-air) add features and/or optimize the performance of deployed devices.
Design, build, maintain efficient and reliable code. Familiar with Unix environment, Shell scripting, and Git-based source control systems. Basic knowledge to design and develop SW that can simulate the behavior of devices.
Familiar with writing code utilizing concepts from multi-threading, RTOS, OOPS. Experienced using inline functions, volatile keywords, macros, interrupts, virtual, friend functions.
Experienced in setting up central hosting environments such as Azure, ThingsSpeak, Dweet, IBM Watson, Node-Red, Digital Ocean, AWS. Then connect the data stream from IoT edge devices to thus configured remote cloud for IoT analytics, historical data storage, and others.
ML Datasets Creation, Analysis, Processing and Visualization
Usage of D-Tale, Pandas profiling to perform Exploratory Data Analysis (EDA) for a given dataset and generate profile reports that contain quantile statistics, descriptive statistics, most frequent values, histogram, correlations, missing values, file and image analysis, text analysis, etc.
Experienced in manipulating and analyzing data from different sources, followed by theoretical and empirical research to answer questions. Understand, process, and utilize data to build models with novelty, applicability, and practicality in mind.
Experienced in building ML datasets: Recently created a multi-sensor dataset named COVID-away. It contains the recording of accelerometer, gyroscope, barometric pressure & rotation vector data for 2071 dynamic hand-to-face movements, performed with various postures (standing, leaning, slouching, etc.) and wrist orientations (variations in Roll, Pitch, and Yaw).
Frequent usage of Plotly, Seaborn, and Matplotlib data visualization libraries in a research environment to present ML model performance, visual comparison of various analytics results, etc.
Handling multiple development aspects from edge to cloud; clear verbal and written communication; creating release notes, releasing and archiving projects; using change management systems (JIRA); basics of Docker, Matlab, and LabVIEW; experienced using Blackboard learning management system, Qwiklabs, BlueJeans, and CoderPad.
Ensuring design complies with Industrial/relevant standards; technical documentation; ensure all health, safety, environmental and regulatory requirements are met; supporting quality and manufacturing groups. Have assisted with specialized customer requests, provided support, traveled to support on-site if necessary.
Knowledge of PCB manufacturing processes, assembly, and used tools. Experienced in prototyping without using PCBs: After the design phase, I build the first prototype manually. Thus, I learned to handle and build prototypes using surface mount technology (SMT), improved hand soldering techniques, reflow soldering, learned PCB etching, drilling, components assembly at home.
My scholarly service to the research community.
ECML PKDD.Sept 2021
Wrote [Proposal] for a half-day tutorial titled 'ML meets IoT: From Theory to Practice', which was accepted at European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD). Tutorial agenda, slides, and code can be found at the [Tutorial Website].
Programme Committee Member
Journals, International Conferences and Workshops.2019 - Present
My role as a Research Paper Reviewer/Programme Committee Member in the below venues is to evaluate submissions based on the quality, completeness, and accuracy of the presented research. Also, provide feedback on the submitted article, suggest improvements and make a recommendation to the editor about whether to accept, reject or request changes to the article.
Taylor and Francis Applied Artificial Intelligence Journal.
Future Generation Computer Systems Journal.
IEEE Access Journal.
International Conference on Artificial Neural Networks (ICANN).
Global IoT Summit (GIoTS).
IEEE World Forum on Internet of Things (WF-IoT).
International Conference on the Internet of Things (IoT).
International Conference on Machine Learning Technologies (ICMLT).
Cross Domain Conference for Machine Learning and Knowledge Extraction (CD-MAKE).
International Conference on Information System and Data Mining (ICISDM).
Embedded and Mobile Deep Learning (ACM MobiSys workshop).
Internet of Things based Health Services and Applications (IoT-HSA).
Universities, Research Institutes.2019 - Present
2021: Delivered a 40 minutes lecture at Newcastle University to cover 'Osmotic Distributed IoT Training' [Pdf Slides].
2021: Delivered a 3 hours session at 'Raksha University, India' to cover the 'Role of ML in IoT Security' [Cert].
2019: Contributed to organizing and presented at the half-day 'Deep Learning with PyTorch' workshop at 'Insight Centre for Data Analytics' [Pics].
Leveraging my contributions to the Edge Computing, IoT, TinyML domains, I assisted Professors, Co-Principal Investigators, Research Fellows in crafting top-quality grants. Below are the local project grants I won by individually writing the proposal.
Nov 2019 - Dec 2020
Liveliness Detection Sub-system for Digital Voice Assistants
Blackstone Launchpad - NUIG
Using the grant, we designed a light-weight Infineon radar-based sub-system and integrated it with Alexa digital voice assistant. Our sub-system enables Alexa to intelligently differentiate live human voices from voices coming out of speakers, thus making Alexa not react to wake word calls and voice commands from non-lively objects.
Dec 2018 - Sept 2019
Assess, Respond, Monitor, Strengthen Glove (ARMS glove) for stroke
Launchpad - NUIG
Post hand paralysis or injury, patients often require lengthy, repeated and therapists supervised clinical training to regain muscular control and function. Using the grant, we built a wearable named Assess, Respond, Monitor, Strengthen (ARMS) glove that facilitates patients to perform various supervised interventions at their convenient place and time without the presence of therapists.
Education and Public Engagement (EPE) Champion - CONFIRM Smart Manufacturing
In 2021, aside from conducting cutting-edge research, I significantly contributed to EPE activities by communicating our scientific breakthroughs to over 2000+ technical audiences as well as the public via tutorials, live demonstrations, interactive posters, and talks. Regardless of baseline knowledge, I aimed to help the audience develop an understanding of enabling technologies for TinyML, IoT, Cyber-Physical Systems, Ubiquitous and Pervasive Computing. Few sessions went more informational based on the audience's needs.
Avoid Touching Your Face - 2nd place at IoT-HSA Workshop
Presented our paper titled "Avoid Touching Your Face: A Hand-to-face 3D Motion Dataset (COVID-away) and Trained Models for Smartwatches" in the IoT-HSA workshop and won the Second place [Link].
Secure Baggage using PSc - Runner-up at OpenGovDataHack
Personal Security (PSc) is a sensor-based embedded system with BLE, which we built to act as a virtual locker for securing the belongings of passengers using public transports. We were the runner-up of NIC-IAMAI #OpenGovDataHack conducted across 7 cities nationwide and qualified for final presentation before Shri Ravi Shankar Prasad, Minister of Electronics & Information Technology, Government of India. The project also got nominated for Tata Consultancy Services (TCS) best project award. Also participated in Smart India Hackathon by The Ministry of Civil Aviation, India [Link].
Smart Portable IoT Vaccine Monitor - 2nd place at Mouser IoT Design Contest
When the environment is not optimal, the efficacy of vaccines is lost, especially when health workers carry vaccines in a portable box during door-to-door polio campaigns. The designed 'Smart Portable IoT Vaccine Monitor' continuously monitors the vaccines using multiple sensors and runs local analytics to ensure vaccine efficacy is preserved. The timely alerts from our device prevent administering less potent vaccines during campaigns. This proposed concept and prototype won second place at Mouser IoT Design Contest, National level, India [Link].
Gesture Control Glove - Finalist at Atmel Embedded Design Contest
During seminars/presentations, to provide a seamless user-machine interaction, we built a wireless sensor-based wearable that helps the presenter achieve improved synchronization while performing presentation control tasks such as window switching, scrolling, slide navigation, audio-video controls, etc. This proposed 'Gesture Control Glove' wearable was the finalist at the national level (India) 'Embedded Design Contest' organized by 'Atmel Semiconductor' [Cert].
Compliance Training - ARM
Completed the following training courses that are intended to provide key guidance on how to conduct oneself when working in a organization [Cert].
Introduction to Patents, Processes for Third Party IP, Open Source and Standards.
Active Shooter or Armed Threat.
Introduction to Open Source Software and Licenses.
Introduction to Intellectual Property Law.
Security Awareness, AI Ethics Awareness Training for Engineering.
Code of Conduct Trade Compliance, Training and Annual Policy Acceptances.
Research Integrity - Engineering and Technology - Epigeum, Oxford University Press
Planning your research, Research with human participants, Managing and protecting interests, Financial interests and intellectual property, Research record, Research communication, Data interpretation and presentation, Case studies and advice: Pressure to publish, Advocacy [Cert].
Health Research and Data Protection Training - NUI Galway
Ethics and DP, Health research Legal framework, mandatory DPIA, PBD&D, Health research projects before GDPR, GDPR, Personal data & SCPD, Data transfers, HRR overview, What is health research, Suitable & specific measures, Legal Basis, Explicit consent, HRCDC, Anonymisation & Pseudonymisation & Techniques, Combining datasets, software & MDR, Explanations, Covid-19 & Health Research.
GDPR Introductory Training - NUI Galway
GDPR Principles Privacy by Design & Default; Material Scope; Territorial Scope; Data flows; Personal Data and Special Category Personal Data; Data Subject Rights; Lawful processing; Fairness & Transparency; Automated decision making & Profiling; Derogations and purpose limitations for research; Anonymisation & pseudonymisation; Combining Datasets; Security; Disclosure, Data Breach & Penalties; Roles & Relationships; When a DPIA is required.
Academic Publishing and Peer Review - Publons Academy
Communication with editors, author and reviewer biases, evaluating data in tables and figures in the results section, structure and effectively communicate your constructive review.
English for Academic Purposes (EAP) workshop - English Language Centre, NUI Galway
Applying academic conventions in the sections of a research paper. Applying appropriate language and communication strategies to academic tasks (describing trends, analyzing data, evaluating the arguments of others, supporting arguments with testimony and evidence, linking ideas logically and coherently). Academic writing (topic sentences, unity, internal cohesiveness, nominalization, sentence structure, strategies for summarising and synthesis).
Google IT Automation with Python (Professional Certificate) - Google via Coursera
Configuration Management and the Cloud (Automation at Scale, Basic Monitoring & Alerting, Cloud Computing, Using Puppet) [Cert]. Crash Course on Python [Cert]. Using Python to Interact with the OS, (Regular Expression (REGEX), Automating System Administration Tasks with Python, Bash Scripting) [Cert]. Introduction to Git and GitHub (Using Git, Version Control Systems, Interacting with GitHub, Reverting Changes, Creating Pull Requests) [Cert]. Troubleshooting and Debugging Techniques (Improving Software Performance, Managing Scarce Resources, Advanced Troubleshooting, Understanding Errors, Finding the Root Cause of a Problem) [Cert]. Automating Real-World Tasks with Python (Serialization, Building a Solution, Creating and Translating Media Files, Interacting with Web Services)[Cert]. Final Professional Certificate [Cert].
Industrial IoT Markets and Security - University of Colorado Boulder via Coursera
Automation deployment, IIoT software and services market, Real-time operating system for an IIoT node, Networking, wireless communication providers and protocols, Network Functions Virtualization and Software Defined Networks, Security solutions for end-node type devices [Cert].
Open Source and the 5G Transition - LinuxFoundationX via edX
Open 5G network, Standards & software, Integrating 5G into business strategy, Considerations & going forward [Cert].
Applied AI with DeepLearning - IBM via Coursera
Deep Learning Frameworks (Keras, TensorFlow, SystemML & DeepLearning4J), DeepLearning Applications (Anomaly Detector, Time Series Forecasting, Image classification & Sequence Classification), Scalingand Deployment (IBM Watson Visual Recognition, Tasks in ApacheSpark using DL4J & SystemML) [Cert].
Cybersecurity and the Internet of Things - University System of Georgia via Coursera
Organizational Risks in industrial Sector, Application in Smart Grid, Security & Privacy Issues, Interoperability & Securityissues, Connected Home & Community, Consumer Wearables(Wearable Computing, Objective Metrics, Quantified Self) [Cert].
Architecting Smart IoT Devices - EIT Digital via Coursera
Hardware & Software for EmS (MCU, SOC, FPGA, Cache, pipeline & coupling, Sensor Networks, Protocolstacks, Licenses, SensorTag Experiment), RTOS (Real-time Scheduling, Synchronisation and Communication, Device Drivers), System Finalisation (Code Tuning, Security, Realtime & Logical remote debugging, Simulation on host) [Cert].
Smartphone Design Training Program - MediaTek Taiwan
Mobiledesignprocess, Performance testing & tuning, Basic BSP knowledge, Taiwan mobile phone eco-system tours, Real practice in MediaTek, Digital /Analogue /Cellular RF /Wireless Connectivity /Multimedia relative knowledge, Camera/Audio tuning, Power consumption & thermal design, Certification & regulation, Case study (measurement and debugging), Mobile market segmentation & positioning [Cert].
Fundamentals of Digital Image and Video Processing -North western University via Coursera