Researcher tools: code & datasets

A deep learning approach towards the large-scale prediction and analysis of bird acoustics from 100 different bird species

Protecting GANs from membership inference attacks at low cost.

A Microsoft AI for Humanitarian Action study in collaboration with the NLRC 510 global initiative. In this study, we leverage high-resolution satellite imagery to conduct building footprint segmentation and train a classifier to assign each building’s damage severity level via an end-to-end deep learning pipeline. Knowing the damage to individual buildings will enable calculating accurately the number of shelters or most impacted areas by natural disasters required in large-scale disaster incidents such as a hurricane.

Pilot Microsoft Cognitive Services to unlock the strategic value of UN unstructured content by building on AI and semantic technologies.

This project aims to predict the probability of leprosy using skin lesion images and clinical data (as compared to the diagnosis of dermatologists). This model is provided for research and development use only. The model is not intended for use in clinical decision-making or for any other clinical use and the performance of model for clinical use has not been established.

AI for Earth project, in collaboration with the Wildlife Conservation Society Colombia (WCS Colombia) to create up-to-date land cover maps of the Orinoquía region in Colombia. We used a land use and land cover (LULC) map that was manually produced using satellite imagery and field data from 2011-2012 to train a semantic segmentation model for 12 land cover classes.

ShowWhy is an interactive application that guides users through the process of answering a causal question using observational data. In use, it has the potential to empower domain experts (who may not be data scientists) to develop a higher standard of evidence than could be achieved using conventional forms of exploratory data analysis (since correlation does not imply causation). In other words, ShowWhy enables emulation of randomized controlled trials that produce a high standard of…

An implementation of feature disentanglement/domain adaptation methods for training deep learning models from x-ray imagery.

SuperSolver is an algorithm written in Sage/Python that solves the general supersingular isogeny problem using the Delfs-Galbraith algorithm. It accompanies the paper “SuperSolver: accelerating the Delfs-Galbraith algorithm with fast subfield root detection”, by Maria Corte-Real Santos, Craig Costello and Jia Shi.

A Python package for generating concise, high-quality summaries of a probability distribution In distribution compression, one aims to accurately summarize a probability distribution $\P$ using a small number of representative points. Near-optimal thinning procedures achieve this goal by sampling $n$ points from a Markov chain and identifying $\sqrt{n}$ points with $O(1/\sqrt{n})$ distributional discrepancy to $\P$. Unfortunately, these same algorithms suffer from quadratic or super-quadratic runtime in the sample size $n$. To address this deficiency, we…