Link here multiplix.net

MultipliX is a website that allows users to customize mental math tests and compete with others. This takes inspiration from the website MonkeyType, which allows users to customize their experience and their touch-typing tests Websites that allow you to test mental math ability tend to be simpler and not as advanced as touch-typing sites typically are.

MultipliX uses Django with a PostgreSQL database on the backend. The frontend uses Bootstrap and vanilla JavaScript for user responsiveness and single-page views.

MultipliX is hosted on an AWS EC2 instance using Gunicorn and Nginx, following this tutorial. Https functionality is implented with Certbot.

Features include...

Responsive input handling on test settings page and login/registration using JavaScript.

Implementing SQL databases to allow user's tests to compete on leaderboards.

Responsive spacing for mobile users.

This is the site you are currently viewing.

This site is a responsive static single-page application utilizing HTML, CSS(Bootstrap), and vanilla JavaScript that is hosted with Github pages.

This site is used as a personal website to display information about myself.

More information on the About page.

Using the software JFLAP, this project models a Turing machine that can add ternary numbers.

By using carry propogation and handling edge cases, this machine can effectively add numbers of different lengths.

This project also includes a Python script for generating and testing cases.

This project is from Binghamton University's CS 373, Formal Language & Automata Theory.

This project uses C++ to model the hit rates of different types of caches, interactions of these caches in multi-level systems, the effectiveness of least recently used mechanisms, and different write policies for store instruction in multi-tiered systems.

Direct-mapped, set associative, and fully-associative caches are modeled. These cache types are also applied as an L1 cache in conjuction with an L2 cache.

Some of these implementations require a LRU mechanism, and this project implements a genuine LRU mechanism with an array of priority and a hot and cold bit LRU approximation.

The impact of write-back and write-through policies in a two level cache is also modeled, showing the efficiency of a write-back policy.

This project's results indicate that higher associativity tends to be more accurate, write-back policies are more efficient, and although hot and cold bit approximations are not as accurate as a true LRU, their results are similar.

This project is from Binghamton University’s CS 320, Advanced Computer Architecture.

This code is 100% original, and the test files and expected outputs were given by the course instructor.

Pre-Processing Methods to Reduce Racial Discrimination of Machine Learning Models in Credit Scoring.

Technical research paper on the discrimination of racial minorities in the US from AI credit-scoring algorithms and why pre-processing methods on training data is a viable solution.

Abstract: Machine learning has replaced traditional statistical methods in credit scoring. Although this has improved the accuracy of credit scoring, benefiting both borrowers and lenders, discrimination in credit scoring against marginalized groups in the US still exists. This discrimination results from patterns in observed features among racial minorities, causing racial discrimination without actually accounting for race. Compared to post-processing methods, typically used to increase the accuracy of machine learning algorithms by using multiple individual models, feature reweighing, a pre-processing method, is more effective at reducing the discrimination in credit scoring models, thus promoting equal financial access. Pre-processing methods have successfully been used to reduce discrimination towards racial minorities in the US with patterns of features that correlate to negative credit decisions. The causes of these patterns in racial minorities in the US can be attributed to discriminatory policies that have marginalized these groups in the past. Policies in the past have limited financial access to certain communities, which have influenced specific features that are accounted for in credit scoring, such as debt-to-income and loan-to-value ratios. Thus, discrimination in credit scoring is a result of past discrimination and needs to be specifically targeted with pre-processing methods on credit data.

This project uses C++ to simulate bimodal, gshare, and tournament branch predictor to evaluate prediction accuracy.

Bimodal branch predictors map addresses to 1-3 bit counter configurations with different sized indexes.

Gshare branch predictors add a layer of complexity to the traditional mapping by index bits by applying an XOR operation with a global history register to get the index of an address.

The tournament branch predictor addes another layer of complexity by indexing each address to a table, where a 2-bit saturating counter chooses between a gshare and bimodal predictor. All three predictors for an address are updated appropriately based on the result.

This project's results show that these added layers of complexity and specificity increase prediction accuracy with the provided test files.

This project is from Binghamton University’s CS 320, Advanced Computer Architecture.

This code is 100% original, and the test files and expected outputs were given by the course instructor.

Network is a single-page Twitter-like social network that allows users to post, follow other users, and interact with posts.

The front-end utilizes JavaScript to interact with posts and CSS to animate elements.

The back-end utilizes Django to create and serve API calls that are used to store and represent post and user data, as well as providing information about the current user and session.

Network is project 4 from Harvard University's CS50W, and basic login/register/logout functions were included in the starter code.

This project is an extension of project 4, Network from Harvard University's CS50W

Mail is a single-page email application that allows users to send emails to each other through API calls.

Users send and interact with emails through the front-end and JavaScript, which makes API calls with Django. Users can send, archive, and reply to emails.



This project is an extension of project 3, Mail from Harvard University's CS50W

This program uses C to take a double floating point value and use bit twiddling to store it into a floatx value with user-specified exponent and fraction bit sizes.

The floatx value will be stored in the right most bits of a double value, meaning that the total bits(exponent bits, fraction bits, and sign bit) have a maximum of 64. There must be at least 1 fraction bit and 1 exponent bit, and the minimum number of total bits is 3.

According to the IEEE standard, a double point floating values has 1 sign bit, 11 exponent bits, and 52 fraction bits. Some values may need more magnitude, i.e. more exponent bits, and less precision, i.e. less fraction bits, or vice versa. A value simply may not need the extent of precision and size given by a standard single or double floating point.

This project is an extension of a project from Binghamton University's CS220

This is a report-style linear regression analysis on the Hitters dataset from R's ISLR package.

This report includes a full exploratory analysis of the data and uses regression analysis to transform and create a regression model to accurately fit the data.



This project is the final project from Binghamton University's MATH329