ben-sanati
PhD Student @ University of Edinburgh EPSRC CDT in Machine Learning Systems
University of EdinburghLondon
Pinned Repositories
3d-localisation-and-mapping
This repository is part of an Innovate UK AKT project, a collaboration between the University of Southampton and an industry client. It involved dataset curation, mobile self-localisation, 3D mapping and object detection.
ben-sanati
Class-Granular-Classifications
This project investigates the accuracy-specificity trade-off of early-exiting dynamic DNNs with two novel architectures: Gran-HBN and AG-HBN.
COMP6237-Data-Mining-Project
A project using economic indicators to predict the performance of the UK luxury fashion sector. A dataset of macroeconomic indicators is created and various time-series models are trained and tested. The Prophet model was found to be the best performing model, achieving a MAPE of 11.36%.
Computational-Finance-CW
This report explores time series analysis, algorithmic trading, and portfolio optimization. It implements an ARIMA model for forecasting, compares pairs trading strategies on Tesco and Pershing Square Holdings, and derives an efficient portfolio using five stocks from the FTSE 100 index for comparison against a 1/n baseline portfolio.
Deep-Learning-Reproducibility-Challenge
A reproducibility project on the paper 'Gradient Descent: The Ultimate Optimizer'. The paper introduces hyperoptimisers that use automatic differentiation to compute optimal hyperparameter values during training. The findings support the claims proposed by the paper and provide further insight into additional features of interest.
HackAI-22
AI Hackathon arranged by USAIS and hosted by Cirium at the University of Southampton. The aim of this hackathon was to process data about organized events and online flight query volumes to determine the events that lead to a spike in flight requirements. This would be used to inform airlines about when flights could be scheduled optimally.
HackAI-23
AI Hackathon arranged by USAIS and hosted by Cirium at the University of Southampton. The aim was to forecast load-factor forecasts for 50 airlines, system-wide (how full on average the planes are). The top 6 teams presented their approach and findings to a panel of judges. The contributors performance earned us a summer internship at Cirium.
MIPS-Processor-Simulator
This project implements a MIPS processor simulator in C++. It interprets a limited set of MIPS assembly, enabling the implementation of a functional processor. The system features an automated assembly-machine code translator and offers modes for testing and execution. Specialized modules handle operations like arithmetic, multiplexing, etc.
AutoSign
AutoSign provides an improvement upon manual train sign inspection by identifying non-conformances with the use of a mobile app and machine learning techniques. Comparative evaluation demonstrated AutoSign's superiority in inspection time, skill requirements, and reliability, proving its feasibility for commercial development.
ben-sanati's Repositories
ben-sanati/3d-localisation-and-mapping
This repository is part of an Innovate UK AKT project, a collaboration between the University of Southampton and an industry client. It involved dataset curation, mobile self-localisation, 3D mapping and object detection.
ben-sanati/Computational-Finance-CW
This report explores time series analysis, algorithmic trading, and portfolio optimization. It implements an ARIMA model for forecasting, compares pairs trading strategies on Tesco and Pershing Square Holdings, and derives an efficient portfolio using five stocks from the FTSE 100 index for comparison against a 1/n baseline portfolio.
ben-sanati/ben-sanati
ben-sanati/Class-Granular-Classifications
This project investigates the accuracy-specificity trade-off of early-exiting dynamic DNNs with two novel architectures: Gran-HBN and AG-HBN.
ben-sanati/COMP6237-Data-Mining-Project
A project using economic indicators to predict the performance of the UK luxury fashion sector. A dataset of macroeconomic indicators is created and various time-series models are trained and tested. The Prophet model was found to be the best performing model, achieving a MAPE of 11.36%.
ben-sanati/Deep-Learning-Reproducibility-Challenge
A reproducibility project on the paper 'Gradient Descent: The Ultimate Optimizer'. The paper introduces hyperoptimisers that use automatic differentiation to compute optimal hyperparameter values during training. The findings support the claims proposed by the paper and provide further insight into additional features of interest.
ben-sanati/HackAI-22
AI Hackathon arranged by USAIS and hosted by Cirium at the University of Southampton. The aim of this hackathon was to process data about organized events and online flight query volumes to determine the events that lead to a spike in flight requirements. This would be used to inform airlines about when flights could be scheduled optimally.
ben-sanati/HackAI-23
AI Hackathon arranged by USAIS and hosted by Cirium at the University of Southampton. The aim was to forecast load-factor forecasts for 50 airlines, system-wide (how full on average the planes are). The top 6 teams presented their approach and findings to a panel of judges. The contributors performance earned us a summer internship at Cirium.
ben-sanati/MIPS-Processor-Simulator
This project implements a MIPS processor simulator in C++. It interprets a limited set of MIPS assembly, enabling the implementation of a functional processor. The system features an automated assembly-machine code translator and offers modes for testing and execution. Specialized modules handle operations like arithmetic, multiplexing, etc.
ben-sanati/scrapy-intro
A Scrapy introductory project following The Scrapy Playbook online series. This series creates a book scraper that scrapes the following webpage: https://books.toscrape.com/.
ben-sanati/ViT-MOT
During my summer UG research internship, I explored the use of vision transformers (ViTs) for object detection, focusing on mitigating the computational requirements of transformer models. This project successfully achieved multiple objectives and identified crucial features for future advancements in object detection using ViTs.
ben-sanati/Year-2-Group-Project
Our Year 2 University project, Abor.io, promotes remote employee well-being during COVID-19. It incentivizes group workouts, fosters community, and maintains relationships. By encouraging simultaneous activity and offering rewards, Abor.io improves mental and physical health while ensuring productivity.