ABOUT ME

Prognosis of Vulnerability Towards Drug Addiction through Supervised Machine Learning

• Title
• Abstract

Prognosis of Vulnerability Towards Drug Addiction through Supervised Machine Learning

Fahim Faisal;Arif Shahriar; Sohan Uddin Mahmud; Amitabha Chakrabarti; Md Golam Rabiul Alam;
Thesis Archived Pre-print , 2019.
[paper], [arXiv], [youtube].

Addiction has become a major problem world wide, and the main problem behind its addictive nature is that the person cannot identify how gradually he is becoming more dependent on narcotics / other substances. One solution to this problem is that identify a person that he/she is prone to addiction before getting in the chronic stage so that he can be brought under proper treatment procedure. In this era of machine learning, utilizing predictive models can help us identify key traits that lead to addiction using feature importance and can come up with models that can identify whether a person is getting addicted or not. Coming up with such a solution will require a data set with enriched features because recent machine learning approaches are data-driven, and we tried to come up with a novel dataset that contains - social, legal, marital issues etc. So, early reconnaissance of substance use disorder may help to take primordial measures for its remedy. There are many inequalities between an addicted and a non-addicted person in terms of their social and family conduct. In this study, the writers have tried to figure out a person's traits that are connected to his or her social and family life as well as health problems that may show his or her exposure to drug abuse. This research was carried out on the scope of the citizens of Bangladesh and an age group limited to 15 to 40 years. A primary data set was constructed following some international scales (WHO,ASI, etc.) which include 498 samples.For drug-addicted person's information, some rehabilitation centers of Dhaka have considered this. For non-addicted persons data, they also communicated with people of different ages from various colleges and universities. 498 samples were collected where one sample consisted of 60 features were trained and tested by a supervised machine learning approach. The reliability of the data set was validated by Cronbachs Alpha Nominal Test. 5 algorithms were incorporated, including Neural Network, Deep Belief Network, Random Forest, XG-Booster, etc. and their results were compared. Among the algorithms, XGB came up with the highest accuracy of 95.20%, and Logistic Regression delivered the least, which is approximately 88%. To select important features mRMR, Chi-square, and Principle Component Analysis techniques were used. Another approach of embedded voting was incorporated to identify key features using Gradient Booster, Support Vector Regression and Random Forest algorithm and Pearson's correlation. From feature selection, the key features of an addicted persons behavior that were influential for their drug abuse have been identified. This research will help people understand whether or not a person will be vulnerable to addiction based on their health conditions and social and family activity.

A Brief Review of Responsible AI and Socially Responsible Investment in Financial and Stock Trading

• Title
• Abstract

A Brief Review of Responsible AI and Socially Responsible Investment in Financial and Stock Trading

Ullah, A. K. M. Amanat; Sultana, Samiha; Faisal, Fahim; Rahi, Md. Muzahidul Islam; Alam, Md. Ashraful;Alam, Md. Golam Rabiul;
Pre-print in TechRxiv , 2019.
[paper], [arXiv], [youtube].

Automated trading is used in most of the major markets of our world. In order to ensure sustainable development, incorporating ethical and socially responsible ideas while designing these Artificial Intelligence (AI) systems has become a necessity. Both the industry and the academia are working towards Responsible AI, which can make Socially Responsible Investments (SRI). This paper reviews the research on SRI investment in the financial sector and evaluates these methods, which can help find future research directions in Computational Finance. This survey looks at the machine learning techniques used for ethical decision-making while stock or forex trading, which will benefit any further research work on Responsible AI in Finance

Privacy-Preserving Learning via Data and Knowledge Distillation

Abstract—In the current era of data science, deep learning, computer vision and image analysis have become ubiquitous across various sectors, ranging from government agencies and large corporations to small end devices, due to their ability to simplify people’s lives. However, the widespread use of sensitive image data and the high memorization capacity of deep learning present significant privacy risks. Now, a simple Google search can yield numerous images of a person, and the knowledge that a specific patient’s record was utilized for training a specific model associated with a disease may reveal the patient’s ailment, potentially leading to membership privacy leakage and other advanced attacks in the future. Furthermore, these unprotected models may also suffer from poor generalization due to this overfitting to train data. Previous state-of-the-art methods like differential privacy (DP) and regularizer-based defenses comrpromised functionality, i.e., task accuracy, to preserve privacy. Such an imbalanced trade-off raises concerns about the practicability of such defenses. Other existing knowledge-transfer-based methods either reuse private data or require more public data, which could compromise privacy and may not be viable in certain domains. To address these challenges, where membership privacy is of utmost importance and utility cannot be compromised, we propose a novel collaborative distillation approach that transfers the private model’s knowledge based on a minimal amount of distilled synthetic data, leading to a compact private model in an end-to-end fashion. Empirically, our proposed method guarantees superior performance compared to most advanced models currently in use, increasing utility by almost 8%, 34%, and 6% for CIFAR-10, CIFAR-100, and MNIST, respectively. The utility resembles non-private counterparts almost closely while maintaining a respectable level of membership privacy leakage of 50-53.5%, despite employing a smaller model with 50% fewer parameters.