|
/
|
<svg width="99" height="31" viewbox="0 0 99 31" fill="none" xmlns="http://www.w3.org/2000/svg" class="LogoHeader_iconResearch__HfDAF"><path d="M3.34 6.16v18.39h3.2v-7.07h1.58l4.08 7.07h3.48l-4.51-7.75c2.3-.72 3.93-2.36 3.93-5.14 0-3.46-2.69-5.5-6.48-5.5H3.34zM8.33 15H6.55V8.65H8.4c2.3 0 3.47.9 3.47 3.01 0 2.17-1.3 3.34-3.55 3.34zm18.76 8.65v-2.57a8.07 8.07 0 01-4.16 1.24c-2.31 0-3.19-1.08-3.32-3.3h7.61v-1.67c0-4.63-2.04-6.37-5.2-6.37-3.85 0-5.68 2.93-5.68 6.95 0 4.63 2.29 6.88 6.32 6.88 2.02 0 3.5-.53 4.43-1.16zm-5.12-10.19c1.57 0 2.04 1.3 2.04 2.97v.26h-4.4c.08-2.12.85-3.23 2.36-3.23zm11.5 11.35c3.12 0 5.17-1.3 5.17-4.44 0-2.33-1.62-3.55-4.3-3.92-1.33-.18-2.44-.34-2.44-1.61 0-1 .8-1.46 2.28-1.46 2.04 0 3.34.74 3.55.9v-2.56s-1.24-.74-3.87-.74c-3.24 0-5.04 1.69-5.04 4 0 2.35 1.35 3.56 3.95 3.9 1.83.24 2.81.75 2.81 1.91 0 1.03-.82 1.53-2.62 1.53-2.31 0-3.8-1.19-4.06-1.32v2.57s1.48 1.24 4.56 1.24zm17.38-1.16v-2.57a8.07 8.07 0 01-4.17 1.24c-2.31 0-3.19-1.08-3.32-3.3h7.62v-1.67c0-4.63-2.04-6.37-5.2-6.37-3.85 0-5.68 2.93-5.68 6.95 0 4.63 2.28 6.88 6.31 6.88 2.02 0 3.5-.53 4.44-1.16zm-5.13-10.19c1.57 0 2.05 1.3 2.05 2.97v.26h-4.4c.07-2.12.84-3.23 2.35-3.23zm17.36 2.01c0-3.22-1.64-4.44-4.99-4.44-2.1 0-3.74.66-4.7 1.22v2.62a8.29 8.29 0 014.33-1.33c1.51 0 2.2.53 2.2 1.96v.74h-.5c-4.83 0-6.98 1.59-6.98 4.29s1.64 4.2 4.09 4.2c1.85 0 2.65-.6 3.26-1.24h.13c.03.34.14.8.24 1.06h3.08c-.1-1.09-.16-2.17-.16-3.26v-5.82zm-3.16 5.9c-.4.58-1.14 1.06-2.25 1.06-1.33 0-2-.74-2-1.88 0-1.5 1.1-2.06 3.8-2.06h.45v2.88zm8.87-5.13a4.07 4.07 0 014.11-2.33v-2.96c-1.8.1-3.29 1.2-4.03 2.94h-.08l-.08-2.65h-3.08v13.3h3.16v-8.3zm11.04 8.57c1.46 0 2.5-.26 3.26-.82v-2.56c-.8.55-1.75.9-3.08.9-2.25 0-3.18-1.75-3.18-4.5 0-2.89 1.14-4.37 3.21-4.37 1.22 0 2.42.43 3.05.82v-2.67a7.38 7.38 0 00-3.4-.63c-4.03 0-6.13 2.88-6.13 6.93 0 4.44 2.05 6.9 6.27 6.9zM88.37 15a3.46 3.46 0 012.66-1.32c1.11 0 1.62.47 1.62 1.48v9.39h3.16v-9.71c0-2.65-1.07-3.81-3.64-3.81-1.88 0-3 .69-3.64 1.32h-.16V6.16h-3.15v18.39h3.15V15z" fill="currentColor"></path></svg><span class="LogoHeader_title__IpbAJ">Yandex Research</span>
|
|
|
/about
|
About
|
|
|
/research-areas
|
Research areas
|
|
|
/blog
|
Blog
|
|
|
/publications
|
Publications
|
|
|
/join
|
Join
|
|
|
/
|
Home
|
|
|
/research-areas/tabular-data
|
<h3 class="ResearchAreasListItem_title__mgWzt">Tabular data<svg viewbox="0 0 12 20" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M2.167 1.833L10.333 10l-8.166 8.167" stroke="currentColor" stroke-width="2" stroke-linecap="square" stroke-linejoin="round"></path></svg></h3><div class="ResearchAreasListItem_text__hhqA3">Tabular data involves two-dimensional tables with objects (rows) and features (columns), which are used in numerous applied tasks such as classification, regression, ranking and many others.</div><div class="ResearchAreasListItem_parameters__dekZR"><div class="ResearchAreasListItem_parametersItem__DqKuK">11<!-- --> <!-- -->publications</div><div class="ResearchAreasListItem_parametersItem__DqKuK">2<!-- --> <!-- -->posts</div><div class="ResearchAreasListItem_parametersItem__DqKuK">2<!-- --> <!-- -->datasets</div></div>
|
|
|
/research-areas/large-scale-machine-learning
|
<h3 class="ResearchAreasListItem_title__mgWzt">Large-scale machine learning<svg viewbox="0 0 12 20" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M2.167 1.833L10.333 10l-8.166 8.167" stroke="currentColor" stroke-width="2" stroke-linecap="square" stroke-linejoin="round"></path></svg></h3><div class="ResearchAreasListItem_text__hhqA3">Today, training most powerful models often takes significant resources. Our research aims to make large-scale training more efficient and accessible to the entire machine learning community.</div><div class="ResearchAreasListItem_parameters__dekZR"><div class="ResearchAreasListItem_parametersItem__DqKuK">18<!-- --> <!-- -->publications</div><div class="ResearchAreasListItem_parametersItem__DqKuK">2<!-- --> <!-- -->posts</div></div>
|
|
|
/research-areas/generative-models
|
<h3 class="ResearchAreasListItem_title__mgWzt">Generative models<svg viewbox="0 0 12 20" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M2.167 1.833L10.333 10l-8.166 8.167" stroke="currentColor" stroke-width="2" stroke-linecap="square" stroke-linejoin="round"></path></svg></h3><div class="ResearchAreasListItem_text__hhqA3">Generative models in computer vision are powerful tool for various applications.</div><div class="ResearchAreasListItem_parameters__dekZR"><div class="ResearchAreasListItem_parametersItem__DqKuK">21<!-- --> <!-- -->publication</div><div class="ResearchAreasListItem_parametersItem__DqKuK">6<!-- --> <!-- -->posts</div></div>
|
|
|
/research-areas/graph-machine-learning
|
<h3 class="ResearchAreasListItem_title__mgWzt">Graph machine learning<svg viewbox="0 0 12 20" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M2.167 1.833L10.333 10l-8.166 8.167" stroke="currentColor" stroke-width="2" stroke-linecap="square" stroke-linejoin="round"></path></svg></h3><div class="ResearchAreasListItem_text__hhqA3">Graphs are a natural way to represent data from various domains such as social networks, molecules, text, code, etc. We develop and analyze algorithms for graph-structured data.</div><div class="ResearchAreasListItem_parameters__dekZR"><div class="ResearchAreasListItem_parametersItem__DqKuK">18<!-- --> <!-- -->publications</div><div class="ResearchAreasListItem_parametersItem__DqKuK">7<!-- --> <!-- -->posts</div><div class="ResearchAreasListItem_parametersItem__DqKuK">2<!-- --> <!-- -->datasets</div></div>
|
|
|
/research-areas/neural-algorithmic-reasoning
|
<h3 class="ResearchAreasListItem_title__mgWzt">Neural algorithmic reasoning<svg viewbox="0 0 12 20" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M2.167 1.833L10.333 10l-8.166 8.167" stroke="currentColor" stroke-width="2" stroke-linecap="square" stroke-linejoin="round"></path></svg></h3><div class="ResearchAreasListItem_text__hhqA3">Algorithmic reasoning focuses on building models that can execute classic algorithms. It allows one to combine the advantages of neural networks with theoretical guarantees of algorithms.</div><div class="ResearchAreasListItem_parameters__dekZR"><div class="ResearchAreasListItem_parametersItem__DqKuK">2<!-- --> <!-- -->publications</div><div class="ResearchAreasListItem_parametersItem__DqKuK">2<!-- --> <!-- -->posts</div></div>
|
|
|
/research-areas/computer-vision
|
<h3 class="ResearchAreasListItem_title__mgWzt">Computer vision<svg viewbox="0 0 12 20" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M2.167 1.833L10.333 10l-8.166 8.167" stroke="currentColor" stroke-width="2" stroke-linecap="square" stroke-linejoin="round"></path></svg></h3><div class="ResearchAreasListItem_text__hhqA3">Yandex Research team regularly contributes to the computer vision research community, mostly in the field of image retrieval and generative modelling.</div><div class="ResearchAreasListItem_parameters__dekZR"><div class="ResearchAreasListItem_parametersItem__DqKuK">43<!-- --> <!-- -->publications</div><div class="ResearchAreasListItem_parametersItem__DqKuK">5<!-- --> <!-- -->posts</div><div class="ResearchAreasListItem_parametersItem__DqKuK">1<!-- --> <!-- -->dataset</div></div>
|
|
|
/research-areas/natural-language-processing
|
<h3 class="ResearchAreasListItem_title__mgWzt">Natural language processing<svg viewbox="0 0 12 20" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M2.167 1.833L10.333 10l-8.166 8.167" stroke="currentColor" stroke-width="2" stroke-linecap="square" stroke-linejoin="round"></path></svg></h3><div class="ResearchAreasListItem_text__hhqA3">Language is one of the key forms of communication. We study methods of language representation and understanding to simplify human-computer interactions.</div><div class="ResearchAreasListItem_parameters__dekZR"><div class="ResearchAreasListItem_parametersItem__DqKuK">36<!-- --> <!-- -->publications</div><div class="ResearchAreasListItem_parametersItem__DqKuK">3<!-- --> <!-- -->posts</div><div class="ResearchAreasListItem_parametersItem__DqKuK">2<!-- --> <!-- -->datasets</div></div>
|
|
|
/research-areas/machine-learning-theory
|
<h3 class="ResearchAreasListItem_title__mgWzt">Machine learning theory<svg viewbox="0 0 12 20" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M2.167 1.833L10.333 10l-8.166 8.167" stroke="currentColor" stroke-width="2" stroke-linecap="square" stroke-linejoin="round"></path></svg></h3><div class="ResearchAreasListItem_text__hhqA3">We study various aspects related to theoretical understanding of ML models and algorithms.</div><div class="ResearchAreasListItem_parameters__dekZR"><div class="ResearchAreasListItem_parametersItem__DqKuK">38<!-- --> <!-- -->publications</div><div class="ResearchAreasListItem_parametersItem__DqKuK">2<!-- --> <!-- -->posts</div></div>
|
|
|
/research-areas/optimization
|
<h3 class="ResearchAreasListItem_title__mgWzt">Optimization<svg viewbox="0 0 12 20" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M2.167 1.833L10.333 10l-8.166 8.167" stroke="currentColor" stroke-width="2" stroke-linecap="square" stroke-linejoin="round"></path></svg></h3><div class="ResearchAreasListItem_text__hhqA3">Most machine learning algorithms build an optimization model and learn its parameters from the given data. Thus, developing effective and efficient optimization methods is of the essence.</div><div class="ResearchAreasListItem_parameters__dekZR"><div class="ResearchAreasListItem_parametersItem__DqKuK">27<!-- --> <!-- -->publications</div></div>
|
|
|
/research-areas/nearest-neighbor-search
|
<h3 class="ResearchAreasListItem_title__mgWzt">Nearest neighbor search<svg viewbox="0 0 12 20" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M2.167 1.833L10.333 10l-8.166 8.167" stroke="currentColor" stroke-width="2" stroke-linecap="square" stroke-linejoin="round"></path></svg></h3><div class="ResearchAreasListItem_text__hhqA3">Nearest neighbor search is a long-standing problem arising in a large number of machine learning applications, such as recommender services, information retrieval, and others.</div><div class="ResearchAreasListItem_parameters__dekZR"><div class="ResearchAreasListItem_parametersItem__DqKuK">15<!-- --> <!-- -->publications</div><div class="ResearchAreasListItem_parametersItem__DqKuK">3<!-- --> <!-- -->posts</div><div class="ResearchAreasListItem_parametersItem__DqKuK">1<!-- --> <!-- -->dataset</div></div>
|
|
|
/research-areas/ranking
|
<h3 class="ResearchAreasListItem_title__mgWzt">Ranking<svg viewbox="0 0 12 20" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M2.167 1.833L10.333 10l-8.166 8.167" stroke="currentColor" stroke-width="2" stroke-linecap="square" stroke-linejoin="round"></path></svg></h3><div class="ResearchAreasListItem_text__hhqA3">Learning to rank is a central problem in information retrieval. The objective is to
rank a given set of items to optimize the overall utility of the list.</div><div class="ResearchAreasListItem_parameters__dekZR"><div class="ResearchAreasListItem_parametersItem__DqKuK">17<!-- --> <!-- -->publications</div></div>
|
|
|
/research-areas/uncertainty-estimation
|
<h3 class="ResearchAreasListItem_title__mgWzt">Uncertainty estimation<svg viewbox="0 0 12 20" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M2.167 1.833L10.333 10l-8.166 8.167" stroke="currentColor" stroke-width="2" stroke-linecap="square" stroke-linejoin="round"></path></svg></h3><div class="ResearchAreasListItem_text__hhqA3">Uncertainty estimation enables detecting when ML models make mistakes. This is of critical importance in high risk machine learning applications, such as autonomous vehicle and medical ML.</div><div class="ResearchAreasListItem_parameters__dekZR"><div class="ResearchAreasListItem_parametersItem__DqKuK">10<!-- --> <!-- -->publications</div><div class="ResearchAreasListItem_parametersItem__DqKuK">3<!-- --> <!-- -->posts</div><div class="ResearchAreasListItem_parametersItem__DqKuK">1<!-- --> <!-- -->dataset</div></div>
|
|
|
/research-areas/gradient-boosting
|
<h3 class="ResearchAreasListItem_title__mgWzt">Gradient boosting<svg viewbox="0 0 12 20" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M2.167 1.833L10.333 10l-8.166 8.167" stroke="currentColor" stroke-width="2" stroke-linecap="square" stroke-linejoin="round"></path></svg></h3><div class="ResearchAreasListItem_text__hhqA3">Gradient boosting iteratively combines weak learners (usually decision trees) to create a stronger model. It achieves state-of-the-art results on tabular data with heterogeneous features.</div><div class="ResearchAreasListItem_parameters__dekZR"><div class="ResearchAreasListItem_parametersItem__DqKuK">13<!-- --> <!-- -->publications</div><div class="ResearchAreasListItem_parametersItem__DqKuK">1<!-- --> <!-- -->post</div></div>
|
|
|
/research-areas/distributional-shift
|
<h3 class="ResearchAreasListItem_title__mgWzt">Distributional shift<svg viewbox="0 0 12 20" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M2.167 1.833L10.333 10l-8.166 8.167" stroke="currentColor" stroke-width="2" stroke-linecap="square" stroke-linejoin="round"></path></svg></h3><div class="ResearchAreasListItem_text__hhqA3">Distributional shift is the mismatch between training and deployment data that is ubiquitous in the real-world. Studying this phenomenon can enable safer and more reliable ML systems.</div><div class="ResearchAreasListItem_parameters__dekZR"><div class="ResearchAreasListItem_parametersItem__DqKuK">6<!-- --> <!-- -->publications</div><div class="ResearchAreasListItem_parametersItem__DqKuK">3<!-- --> <!-- -->posts</div><div class="ResearchAreasListItem_parametersItem__DqKuK">1<!-- --> <!-- -->dataset</div></div>
|
|
|
/research-areas/machine-translation
|
<h3 class="ResearchAreasListItem_title__mgWzt">Machine translation<svg viewbox="0 0 12 20" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M2.167 1.833L10.333 10l-8.166 8.167" stroke="currentColor" stroke-width="2" stroke-linecap="square" stroke-linejoin="round"></path></svg></h3><div class="ResearchAreasListItem_text__hhqA3">Language barriers hinder global communication and access to worldwide knowledge. By improving machine translation systems, we hope to facilitate the exchange of culture and information.</div><div class="ResearchAreasListItem_parameters__dekZR"><div class="ResearchAreasListItem_parametersItem__DqKuK">9<!-- --> <!-- -->publications</div><div class="ResearchAreasListItem_parametersItem__DqKuK">1<!-- --> <!-- -->dataset</div></div>
|
|
|
/research-areas/distributed-ml
|
<h3 class="ResearchAreasListItem_title__mgWzt">Distributed ML<svg viewbox="0 0 12 20" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M2.167 1.833L10.333 10l-8.166 8.167" stroke="currentColor" stroke-width="2" stroke-linecap="square" stroke-linejoin="round"></path></svg></h3><div class="ResearchAreasListItem_text__hhqA3">Training and running large neural networks efficiently across many devices, whether in a GPU cluster or a swarm of poorly connected consumer devices.</div><div class="ResearchAreasListItem_parameters__dekZR"><div class="ResearchAreasListItem_parametersItem__DqKuK">7<!-- --> <!-- -->publications</div><div class="ResearchAreasListItem_parametersItem__DqKuK">1<!-- --> <!-- -->post</div></div>
|
|
|
/research-areas/model-compression
|
<h3 class="ResearchAreasListItem_title__mgWzt">Model compression<svg viewbox="0 0 12 20" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M2.167 1.833L10.333 10l-8.166 8.167" stroke="currentColor" stroke-width="2" stroke-linecap="square" stroke-linejoin="round"></path></svg></h3><div class="ResearchAreasListItem_text__hhqA3">Deep learning models are outgrowing the hardware that runs them. We try to make large models fit on smaller devices through quantization, pruning, factorization, and other means.</div><div class="ResearchAreasListItem_parameters__dekZR"><div class="ResearchAreasListItem_parametersItem__DqKuK">7<!-- --> <!-- -->publications</div></div>
|
|
|
/research-areas/speech-processing
|
<h3 class="ResearchAreasListItem_title__mgWzt">Speech processing<svg viewbox="0 0 12 20" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M2.167 1.833L10.333 10l-8.166 8.167" stroke="currentColor" stroke-width="2" stroke-linecap="square" stroke-linejoin="round"></path></svg></h3><div class="ResearchAreasListItem_text__hhqA3">Speech is an important data modality and relatives to applications such as speech recognition and speech synthesis, which are core technologies in products such as vocal assistants.</div><div class="ResearchAreasListItem_parameters__dekZR"><div class="ResearchAreasListItem_parametersItem__DqKuK">5<!-- --> <!-- -->publications</div></div>
|
|
|
/research-areas/speculative-and-parallel-decoding
|
<h3 class="ResearchAreasListItem_title__mgWzt">Speculative and parallel decoding<svg viewbox="0 0 12 20" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M2.167 1.833L10.333 10l-8.166 8.167" stroke="currentColor" stroke-width="2" stroke-linecap="square" stroke-linejoin="round"></path></svg></h3><div class="ResearchAreasListItem_text__hhqA3">Modern LLMs are autoregressive models that generate one token at a time, which is inefficient on parallel hardware. These works accelerate generation by processing multiple tokens per forward pass.</div><div class="ResearchAreasListItem_parameters__dekZR"><div class="ResearchAreasListItem_parametersItem__DqKuK">4<!-- --> <!-- -->publications</div><div class="ResearchAreasListItem_parametersItem__DqKuK">1<!-- --> <!-- -->post</div></div>
|
|
|
/about
|
About
|
|
|
/research-areas
|
Research areas
|
|
|
/blog
|
Blog
|
|
|
/publications
|
Publications
|
|
|
/join
|
Join
|
|