The Mote in AI's Eye: software engineering with agents

Utkarsh Upadhyay and collaborators designed a machine learning algorithm for scheduling lessons, validated through large-scale randomized experiments showing improved learning and retention, published in Nature Science of Learning. Upadhyay also contributed to SciPy and Python, earning a Mars 2020 Helicopter Contributor badge, and led research on learning to crawl web pages with sub-linear regret guarantees, presented at AAAI 2020.

research research , code https://github.com/musically-ut , blog https://blog.musicallyut.xyz , and make stuff projects . You can find my CV here /docs/utkarsh-upadhyay-cv.pdf . As a follow up to Memorize memorize , I worked with Manuel Gomez-Rodriguez https://people.mpi-sws.org/~manuelgr/ and the creators of Swift Learning App https://swift.ch , Christoph Moser and Graham Lancashire, to design a new algorithm for scheduling lessons Select . We ran large scale randomized experiments to verify that the learning indeed was improved by using the ML based instructions. - "Large-scale randomized experiment reveals machine learning helps people learn and remember more effectively" ~ Nature Science of Learning 2021 ; Open Access Paper. https://www.nature.com/articles/s41539-021-00105-8 - A BehindThePaper blog post https://npjscilearncommunity.nature.com/posts/machine-learning-based-instruction-helps-people-memorize-more-effectively . - Networks-Learning/spaced-selection https://github.com/networks-learning/spaced-selection My contributions to the sparse matrix API were recognized by making me a co-contributor and co-author to the Nature Methods paper describing SciPy , which coincided with the release of version 1.0 of the library. Also, this along with my contribution to python/cpython https://github.com/python/cpython , i.e., the Python programming language, also earned me a badge on GitHub for contributing to the Mars 2020 Helicopter Contributor . I always wanted to put something in space 🤗 With Róbert Busa-Fekete https://scholar.google.com/citations?user=JnNcrZoAAAAJ&hl=en , Wojciech Kotłowski https://www.cs.put.poznan.pl/wkotlowski/ , Dávid Pál https://www.david.palenica.com/ , and Balázs Szörényi https://research.yahoo.com/researchers/bszorenyi , I have looked at hte problem of learning to optimally web-crawl pages while simultaneously learning how often they change. Our conclusions about the properties of the learning algorithm and results about learnability of rates of Poisson processes with partial observability apply to many other problems and scenarios as well. We provide the first sub-linear guarantees for such problems and take the first step in the direction of establishing that given some constraints on the optimization problems e.g. RedQueen redqueen , Memorize memorize which schedule events in continuous time, learning the rates/parameters of the environment while simultaneously optimizing is possible with zero-regret. - "Learning to Crawl" ~ AAAI 2020 ; Paper. https://arxiv.org/abs/1905.12781 With Abir De https://cse.iitkgp.ac.in/~abird/ , Aasish Pappu https://research.yahoo.com/researchers/aasishkp , and Manuel Gomez-Rodriguez https://people.mpi-sws.org/~manuelgr/ , I have uncovered a connection between complexity of online discussions and the notion of sign-rank of matrices. This allows us to determine the complexity of online discussions just by looking at the pattern of upvotes/downvotes cast by users on others' comments; the key insight is using humans as oracles and by-passing the nuances of sarcasm and humor often present in online comments. - "On Complexity of Opinions and Online Discussions" ~ WSDM 2019 ; Paper. https://arxiv.org/abs/1802.06807 - Networks-Learning/discussion-complexity https://github.com/Networks-Learning/discussion-complexity With Abir De https://cse.iitkgp.ac.in/~abird/ and Manuel Gomez-Rodriguez https://people.mpi-sws.org/~manuelgr/ , I have developed a deep reinforcement learning algorithm for controlling agents whose actions are performed, and who receives feedback from the environment, at discrete localized points in continuous real time. This is in contract to the classical RL setup where the actions and rewards feedback are synchronously given to the agent at discrete points in time. - "Deep Reinforcement Learning of Marked Temporal Point Processes" ~ NeurIPS 2018 ; Paper. https://arxiv.org/pdf/1805.09360.pdf - Networks-Learning/tpprl https://github.com/Networks-Learning/tpprl - 3-minute video summary https://www.youtube.com/watch?v=JKSpbL0y5LA With Behzad Tabibian https://www.btabibian.com/ , Abir De https://cse.iitkgp.ac.in/~abird/ , Ali Zarezade https://azarezade.github.io/ , Bernhard Schölkopf https://is.tuebingen.mpg.de/de/people/bs and Manuel Gomez-Rodriguez https://people.mpi-sws.org/~manuelgr/ , I have determined the optimal reviewing schedule to keep knowledge fresh in your memory for optimal recall while minimizing effort spent on learning it. - "Enhancing human learning via spaced repetition optimization" ~ PNAS 2019 ; Paper. https://www.pnas.org/content/early/2019/01/18/1815156116 - Networks-Learning/memorize https://github.com/Networks-Learning/memorize - Webpage with summary https://learning.mpi-sws.org/memorize/ With Ali Zarezade https://azarezade.github.io/ , and Manuel Gomez-Rodriguez https://people.mpi-sws.org/~manuelgr/ , I discovered a simple algorithm for keeping your posts visible on your follower's feeds. - "RedQueen: An Online Algorithm for Smart Broadcasting in Social Networks" Oral presentation at WSDM 2017 ; Paper. https://arxiv.org/abs/1610.05773 - Webpage with demo https://learning.mpi-sws.org/redqueen/ - Networks-Learning/RedQueen https://github.com/Networks-Learning/RedQueen With Isabel Valera and Manuel Gomez-Rodriguez https://people.mpi-sws.org/~manuelgr/ , I am developing models to understand how learning happens on Crowdlearning sites, such as Stack Overflow and Wikipedia. - "On Crowdlearning: How do People Learn in the Wild?", oral presentation at Workshop on Machine Learning for Education at NeurIPS 2016 ; - "Uncovering the dynamics of Crowdlearning and the Value of Knowledge", oral presentation at WSDM 2017 ; Paper. https://arxiv.org/abs/1612.04831 With Nan Du, Hanjun Dai, Rakshit Trivedis, Manuel Gomez-Rodriguez, and Le Song, I developed a model which uses recurrent neural networks to model point processes, yielding impressive predictive results. - "Recurrent marked temporal point processes: Embedding event history to vector", Poster presetned at KDD 2016 ; Paper. https://www.kdd.org/kdd2016/subtopic/view/recurrent-temporal-point-process This project has been sun-setted. All data related to the project including messages sent and rooms created has been deleted. An app for chatting which translates chat messages in real time. You can learn a foreign language while not disrupting communication with your friends. - chanslate.in - musically-ut/chanslate https://github.com/musically-ut/chanslate A Chrome/Firefox extension which makes notable forks of Github repositories, well, notable. Did you thank your NodeJS dependencies today? Changes the uninformative paper ID number in the tab title to the paper title on arXiv. Locate your wallpapers on OSX. Now you can set your wallpapers to randomly change every 30 minutes and weed out the ones you don't like at your leisure. A small Chrome extension which allows you to open a link in Incognito mode using a Key combination and a click. A Tweet-bot which looks for issues with label first-timers-only on Github and tweets about them. Generate sequential filenames for saving program state in a threadsafe way. 100% coverage and tested on all OSes. A Python program for converting pdf slides and annotated text notes into Anki decks. Explaining the working of RedQueen: An Online Algorithm for Smart Broadcasting on Social Networks, with the help of a demonstration. A simple visualisation showing how far different parts of the Zürich city are from the nearest public transport. See how users in different tags ask and answer questions on Stack Overflow. See how many users and upvotes different tags see over time on StackOverflow. $P recognizer https://depts.washington.edu/aimgroup/proj/dollar/pdollar.html wrapped as a bower package and with a Touch screen enabled demo. A JS library to color up all those 8b980c and rgba 146, 66, 136, 0.5 lurking around on HTML pages. A delay proxy which can introduce artificial delays to certain requests. Useful for testing behavior of your web-apps when the API calls or local resources take too long to load. Where you will find me: