My research interests are in the theoretical and practical aspects of machine learning, the principles and algorithms that affect learning and how they are applied to specialized domains in making predictions, especially where there are rapid shifts in observable features as well as how mathematical principles, statistics and probabilistic reasoning can be applied to aid better decision making. I am also interested in researching the frontiers involving the synthesis of deep learning and reinforcement learning (what is now known as deep reinforcement learning), and how they can improve the learning task of an agent interacting in a non-deterministic, non-stationary environment (where exact mathematical analysis are no longer feasible), especially when the states of the environment are only partially observable. I also have research experience in the areas of Computer Vision and Natural Language Processing.

Research

AI for diagnosis of skin lesions

Diagnosis of skin diseases across racial divides

The goal of this research is to develop a diagnostic model to identify pigmented and non-pigmented skin lesions in ethnic skin types. An AI model for the diagnosis of skin lesions will augment the work of dermatologists in providing enhanced skincare services across various demographics, particularly in remote under-served communities and also in others that are under-represented in specialist skincare support. Our approach involves applying techniques from Deep Neural Networks and other methods from Computer Vision to classify skin diseases across racial divides.
Website: https://www.aidermatology.org/

AI vs COVID-19 Initiative: BioMedBERT

Natural Language Processing and Understanding Tools for Biomedical Researchers

BioMedBERT advances the state-of-the-art language models for increased information discoverability from biomedical literature. The number of peer-reviewed papers that are published everyday is dramatically increasing. PubMed reports that more than 1 million biomedical research papers are published each year, this amounts to about two papers per minute. Our work leverages the BERT language model architecture to pre-train a large-scale BioMedical language representation model. Our goal is to develop a model that yields state-of-the-art results when fine-tuned on downstream specialized language tasks.
Website: https://www.aivscovid19.org/

Books

Building Machine Learning and Deep Learning Models on Google Cloud Platform

Published

This book seeks to equip the reader from the ground up with all the essential principles and tools for building learning models. Machine learning and deep learning is rapidly evolving, and often it is overwhelming and confusing for a beginner looking to delve into this field. Many have no clue where to start. This book is a one-stop shop that takes the beginner on a journey to state-of-the-art theoretical understanding and practical mastery without assuming any pre-requisite. Although this book is written with the beginner at heart, it is not ridiculously verbose and repetitive, but written in a direct and succinct manner that will appeal to experts and serve as a refresher for the core concepts.

Ekaba Bisong

• Amazon
• Google Books

Java Laboratory Manual: A Quick Starter Guide

Published

This book serves as an introductory text to Computer Programming for first-year undergraduate students of Computer Science, Mathematics and Statistics at the University of Calabar. Each section ends with an end-to-end programming problem, so students have an early feel for sound software development principles. The overall goal of this book is to present the base paradigms of programming and in particular Object Oriented Programming using Java. Notwithstanding, the concepts presented in this book enables students to develop an understanding of the fundamental principles that underscore most popular programming languages such as C++, Python, and JavaScript.
Available only at the University of Calabar, Calabar, Nigeria

Bisong, E.O., Eteng, I.E., Ele, S.L., Arikpo, I.I., Edim, A.E., Ogban, F.U., and Essien, E.E.

Papers

Optimizing Self-Organizing Lists-on-Lists using Pursuit-Oriented Enhanced Object Partitioning

Recently, researchers have proposed the concept of hierarchical Singly-Linked-Lists on Singly-Linked-Lists (SLLs-on-SLLs), where the primitive elements of the primary list are, in and of themselves, sub-lists. In this paper, we propose that the PEOMA reinforcement scheme can be powerful in learning the probabilistic distribution of the Environment to capture dependent elements within the sub-groups. The research shows that the PEOMA-enhanced SLLs-on-SLLs provide results that are an order of magnitude superior to the "de-facto" MTF and TR schemes used in such Environments with so-called "locality of reference".

O. Ekaba Bisong and B. John Oommen

Optimizing Self-Organizing Lists-on-Lists using Enhanced Object Partitioning

This paper considers the problem of minimizing the cost of data retrieval from the most fundamental data structure, i.e., a Singly-Linked List (SLL) in Non-stationary Environments (NSE) exhibiting so-called "Locality of Reference". We show that SLLs-on-SLLs augmented with the Enhanced Object Migration Automaton (EOMA) minimizes the retrieval cost for elements in NSEs and are superior to the stand-alone MTF and TR schemes, and also superior to the OMA-augmented SLLs-on-SLLs operating in such Environments.

O. Ekaba Bisong and B. John Oommen

Built to Last or Built Too Fast? Evaluating Prediction Models for Build Times

Accepted, Proceedings of the 2017 IEEE/ACM 14th International Conference on Mining Software Repositories (MSR 2017). Buenos Aires, Argentina. 20-21 May 2017. This research focuses on finding a balance between integrating often and keeping developers productive. We propose and analyze models that can predict the build time of a job. Such models can help developers to better manage their time and tasks.

Bisong, Ekaba, Tran, Eric, and Baysal, Olga

• arXiv

Writing Series

Learning in Random Environments:
Theory and Practice

Contents:
1: An Introduction to Reinforcement Learning
2: Reinforcement Learning Framework
3: Dynamic Programming
4: Monte-Carlo Methods
5: Temporal-Difference Learning
6: Policy Gradient Methods
7: The Learning Automata Approach
8: Deep Reinforcement Learning

• Github

Kubeflow for Poets: A Guide to Containerization of the Machine Learning Production Pipeline

Contents:
1: Series Introduction
2: Microservices
3: Docker
4: Kubernetes
5: Kubeflow
6: Kubeflow Pipelines
7: Deploying an End-to-End Machine Learning Solution on Kubeflow Pipelines

• Github

Thesis

LA Solutions to Adaptive Data Structures

Ekaba Bisong,

M.Sc. Thesis, 2018

Seminar/ Workshop

Winter 2019: A 2-Day Workshop to Carleton University IEEE Student Chapter and the Ottawa Medical Physics Institute on Machine Learning Using Google Cloud Platform

The workshop content is arranged in the following order:
- An Introduction to Google Cloud Platform (GCP)
- Steps to set-up an account on GCP
- Working with Google Cloud Datalab
- Using Google Colaboratory
- An Overview of Machine Learning
- Linear Regression
- Logistic Regression
- Introducting Deep Learning
- Large Scale Data transformation with Google Cloud Dataflow
- Training/ Inference with Google Cloud Machine Learning Engine

Winter 2018: A 3-Day Workshop to Carleton University IEEE Chapter on Building Machine Learning and Deep Learning Models on Google Cloud Platform

The workshop contains the following parts:
- Part 1: An Introduction to Google Cloud Platform (GCP)
- Part 2: Steps to set-up an account on GCP
- Part 3: Working with Google Cloud Datalab
- Part 4: Using Google Colaboratory

Side Projects

Google Cloud AutoML for Toxicity Classification

September 2019. This project uses Google Cloud AutoML for Natural Language to develop an end-to-end language toxicity classification model to identify obscene text.

Cloud Run: Dataset Summaries

July 2019. This project uses Cloud Run to run a stateless container that employs Pandas profiling to display the summary statistics of a structured CSV dataset.

Medical Image Classification

May 2019. This project uses Google Cloud AutoML Vision to develop an end-to-end medical image classification model for Pneumonia Detection using Chest X-Ray Images.

Cereal Classifier: Android App

July 2017. This project builds a cereal box image classifier from the Google InceptionV3 model with TensorFlow via transfer learning. The app is trained to identify a select number of cereals.

Credit Card Fraud Detection

March 2017. Credit card fraud detection is the science and art of detecting unusual activity in credit transactions. A significant challenge for credit fraud detection research is the availability of real-world data due to privacy and legal concerns.

Predicting Forest Cover Type

February 2017. This project develops a classification model using XGBoost to accurately classify and predict forest cover type from a set of cartographic features. The data is from the UCI Machine Learning repository.

Built to Last or Built Too Fast?

December 2016. Long build times can be an issue when integrations are frequent. This project is a cubist “minimum-viable-product” PoC model to predict the build time of a job.

Analysis of Agricultural Trends in Nigeria

June 2015. This application displays a barplot of several agricultural trends in Nigeria from 1961 to 2012. The collected data is from the World Bank Country Database.