NeurIPS Meetup Turkey 2020
Together with 55 other institutions from all over the world, inzva held the second official satellite NeurIPS meetup that was attended by 82 AI enthusiasts both from Turkey and abroad.
Due to the Covid-19 outbreak, local meetups, which are designed by NeurIPS to make the conference more accessible for everyone, also went online.
inzva organizes the NeurIPS Meetup with an aim to bring experienced local AI enthusiasts from different universities, who either submitted a paper to the conference or follow the sessions closely, together with those who are ready to dive in-depth into the journey of the academic world of artificial intelligence.
Take a look at the presentations below:
Ali Taylan Cemgil, Boğaziçi University - ‘The Autoencoding Variational Autoencoder‘
Ali Taylan Cemgil started his speech with an introduction to self-supervised representation learning and the concept of adversarial attacks. He then introduced the NeurIPS 2020 spotlight paper he co-authored with his colleagues, “Autoencoding Variational Autoencoder”, which is a variant of variational autoencoders that is robust against adversarial attacks and can also perform self-supervised learning.
Aykut Erdem, Koç University - ‘CRAFT: A Benchmark for Causal Reasoning About Forces and inTeractions’
A team effort consisting of academics, graduate and undergraduate students, Aykut Erdem from KUIS AI Lab presented the team’s recent visual question answering dataset, CRAFT, that requires causal reasoning about physical objects. The dataset contains 38,000 video and question pairs. Their benchmark results on the introduced dataset showed that there is still significant room for improvement until machine learning models truly catch up with human performance. Their paper is accepted to the NeurIPS 2020 Workshop on Shared Visual Representations in Human and Machine Intelligence (SVRHM).
Kemal Öksüz, METU ImageLab, ‘A Ranking-based, Balanced Loss Function Unifying Classification and Localisation in Object Detection’
Kemal Öksüz from METU ImageLab presented the lab’s NeurIPS 2020 spotlight paper that adapts the Localisation-Recall-Precision (LRP) loss proposed earlier to average Localisation-Recall-Precision (aLRP) within the context of object detection tasks. Using aLRP metric as an alternative to average precision (AP) significantly improves the performance of object detection models. As a nice bonus, aLRP also has only one hyperparameter to tune in comparison to state-of-the-art object detectors.
Almira Bağlar, Istanbul Technical University - Uras Mutlu, Boğaziçi University
Discussion on “A Future of Work for the Invisible Workers in A.I.”
Aside from the technical ones, the wide adoption of artificial intelligence systems comes with many social opportunities and challenges; just like every Industrial Revolution, the Fourth Industrial Revolution has also a great impact on the way we work. While Future of Work discussions are traditionally held within the framework of the increased-decreased demand for certain jobs, people like data taggers, who are fueling the supervised learning models are usually left out. Almira Bağlar and Uras Mutlu held a discussion based on Saiph Savage’s invited talk featured at NeurIPS 2020 and talked about unregulated ghost work in AI within the context of Future of Work conversations.
You can see the original presentation here.
Furkan Gürsoy, Boğaziçi University, “Hyperbolic Representation Learning”
Our first presenter from our “Call for Presenters” slot, Furkan Gürsoy from Boğaziçi University talked about an exciting, relatively new topic: Hyperbolic representation learning. He went through the first part of the session by introducing the theory and showing examples of application areas. He then moved onto explaining a selection of papers about hyperbolic representation published at NeurIPS 2020.
Berivan Işık, Stanford University, “Noisy Neural Network Compression”
For our next session of call for presenters, Berivan Işık from Stanford University talked about noisy neural network compression in general and as well as having introduced her and her colleagues’ work “Noisy Neural Network Compression for Analog Storage Devices”, which was accepted to Deep Learning through Information Geometry workshop of NeurIPS 2020. In her presentation, she introduced the notion of neural network compression and presented her own results and insights.
Aspiring to create a platform for bringing together more experienced AI enthusiasts with newcomers, we will continue to democratize AI conferences by collaborating with official meetups initiated by top conferences in the field with an aim to overcome obstacles such as expenses, visa issues, distance, and time limits. Stay connected for more.
inzva is supported by BEV Foundation, an education foundation for the digital native generation which aims to build communities that foster peer-learning and encourage mastery through one-to-one mentorship.
Subscribe to our newsletter here and stay tuned for more hacker-driven activities by following us on social media.