We're looking for talented and motivated post-doctoral researchers in multiple areas of AI and ML. We provide a competitive benefits package, please check our page for details and spread the word!

Given recent advances in learned video prediction, we investigate whether a simple video codec using a pre-trained deep model for next frame prediction based on previously encoded/decoded frames without sending any motion side information can compete with standard video codecs based on block-motion compensation. Frame differences given learned frame predictions are encoded by a standard still-image (intra) codec

Dr. Didem Unat is awarded with an ERC starting grant of 1.5M EUR. First ERC in Computer Science from Turkey! Congratulations on her big achievement.

Najeeb Ahmad, a PhD student of Dr. Didem Unat received the “Best Artifact” award at Euro-Par 2020 conference (24% acceptance rate) held virtually at Warsaw, Poland. Out of the 39 accepted papers, artifacts for 13 papers were accepted and one was chosen for this award. The award was presented during the Best Artifact session held on Thursday 27.08.2020 at 15:00 CET. It was announced in the form of a certificate and 800 Euro cash award.

Congratulations to Dr. Didem Unat for her new EuroHPC grant - Towards Extreme Scale Technologies and Applications.  The action has a total budget of 2.6M Euros with 6 partners in Europe, where Dr. Unat will be leading the Consortium which includes Simula Research Laboratory (Norway), GraphCore Inc. (Norway), INESC-ID (Portugal), LMU Munich (Germany) and Sabanci University (Turkey).

Although speech driven facial animation has been studied extensively in the literature, works focusing on the affectivecontent of the speech are limited. This is mostly due to the scarcity of affective audio-visual data. In this paper, we improve the affectivefacial animation using domain adaptation by partially reducing the data scarcity. We first define a domain adaptation to map affectiveand neutral speech representations to a common latent space in which cross-domain bias is smaller.

Deep Edge Intelligence (DEI) has a tremendous potential to democratize Artificial Intelligence and address challenges caused by the rapid data and device proliferation

We’re excited to share our pre-trained ALBERT models for Arabic! Trained on a corpus of ~4.4B words, these models are the first to use sentence piece tokenization for Arabic. Thanks to @Google for providing TPUs for the training and for @huggingface to host the models.