This is the portfolio representation of Md. Mahmudul Haque. Github handle "alcatraz47"
Who am I?: I am a Data Science practitioner and have been researching in this domain for about 5 and half years. I have a little contribution to this field. I am mostly interested in Natural Language Processing though my first job as well as my capstone is on Computer Vision. I have publication on textual NLP and reasearch experience on voice based NLP.
Travelling, Listening songs, Reading fiction books, and Watching motor sports specially Formula 1
Current:
Past:
CGPA: shush!!!…. I am an NSUer!
Text Summarization on news data of financial SMEs in Europe.
Social media bot-based marketing for news distribution:
FaceNext Web and Mobile App: Access controlling system based on face recognition.
Street Visitor: Developing pipelines and predictive analysis of workers GPS locating devices.
ETL on Public Transport Data: ETL on the public transport data for report generation purpose of each day.
Survello Web: Computer vision-based surveillance system.
The proposed arrangement of this undertaking incorporates order of the passages utilising its temperament. Every one of the sections is self-described, and the number of words in those self-described passages vary from more than 100 to under 4200. The passages are classified utilising three classifications which are: ‘‘Work Stress”, ‘‘Bullying” in both social and digital world, and ultimately ‘‘Sexual Harassment” in public activity and digital world. Artificial neural network paragraph vectors: a distributed bag of words and distributed memory were utilised to get the features of each passage and later on to group them a few information mining strategies were employed, and these are: decision tree, k nearest neighbours, Gaussian naive Bayes, and logistic regression. The exactness of every calculation lied between 70\% to 94\% in the validation set. The best model gave 77.46\% F1 score in test sets.
Technical Report Link: ResearchGate
Skin Cancer Detection Using Deep Neural Networks(Paper on Process) The sole purpose of this project was to detect the type/category of skin cancer from the given pigmented skin lesion. The types are Actinic keratoses and intraepithelial carcinoma / Bowen’s disease (akiec), basal cell carcinoma (bcc), benign keratosis-like lesions (solar lentigines / seborrheic keratoses and lichen-planus like keratoses, bkl), dermatofibroma (df), melanoma (mel), melanocytic nevi (nv) and vascular lesions (angiomas, angiokeratomas, pyogenic granulomas and haemorrhage, vasc). The dataset was collected from Kaggle’s Skin Cancer MNIST: HAM10000 dataset. The dataset was split into 3 partitions: Training(70%), Validation(20%), and lastly Testing(10%). At first, the dataset was normalised and then it was segmented according to the area. Then data generator was used to provide more variation during training time in batch. The whole project was done using the following algorithms: Custom CNN, Custom RESNET32, Pre-Trained Resnet50, Pre-Trained Resnet101. The highest accuracy was about 93% in Resnet101.
Liveness/Presentation Error Detection The objective of this project was to catch whether a/some person(s) is really present in a face recognition system or not. The dataset was collected manually by me and my friends. The dataset was directly fed into custom CNN and one Resnet50 algorithm without further preprocessing except resizing. Accuracy on the training set was about 99%.
Bangla Handwritten Digit Recognition I have used a dataset from Kaggle and used some image processing before jumping into the project. At first, I used Random Forest Regression Classifier to detect and recognise the digits. Later on, I have used a Convolutional Neural Network to predict on the digits. Email Spam Classifier using Support Vector Machine(SVM): Here, I used SVM(linear classifier and Gaussian Kernel) to detect whether an email is a spam or not. For this project, I used the labelled dataset of Coursera’s Machine Learning program and also borrowed some optimisation algorithm from it.
Compressing Image Using Clustering Algorithm Here, I used one picture of mine and my friend to compress. At first, I lowered the dimension of it and then used K-means Clustering algorithm into it. Here, I borrowed the optimisation algorithm of Coursera’s Machine Learning program to optimise the parameters of my algorithm.
8th at Team Contest of NeurIPS AutoDL challenges (Auto Speech Challenge) co-hosted by Google, Cha-Learn, and 4paradigm.
Md. Rakib Saleh CEO, Unatitech Email: rakib@unatitech.se
DR. MOHAMMAD RASHEDUR RAHMAN Professor & Graduate Coordinator PhD in Computer Science, University of Calgary, Canada MS in Computer Science, University of Manitoba, Canada BS in Computer Science and Engineering, BUET, Bangladesh.