SAIT collaborates with talented people around the world to keep abreast of state-of-the-art technologies, drive new trends, and create opportunities. Through Open Innovation, SAIT's global labs located in San Jose, Pasadena, Boston, Montreal, Yokohama, Beijing, Xian, Bengaluru, Moscow, Kyiv, Warsaw, and London work together and cooperate with regional universities, research labs, and companies to develop innovative technologies for all.
Global Labs
San Jose Lab
- Supercomputer
- Next-gen Processor
- Tech Scouting
Pasadena Lab
- Optical Interconnect
Boston Lab
- Materials Design
United States
Overview
SAIT officially established SAIT-America in May of 2011, with offices in San Jose, Boston, and Pasadena. SAIT-America is strategically located near cutting-edge R&D activities of prestigious academic institutions, renowned companies, and new exciting startups with a purpose to lead ground-breaking research. SAIT-America’s mission is to plant the seeds of innovative technologies that will grow and yield next-generation Samsung products within 5 to 10 years. Our aim is to achieve this goal by collaborating and working together with the most creative minds in the world.
Research Areas & Core Activities
SAIT-America is composed of 4 research labs with an Open Innovation team. Systems Architecture Lab (SAL) is researching HPC and AI Systems with a focus on overcoming the memory and communication challenges within HPC and AI applications. By
taking a holistic, system-level approach to these obstacles, SAL will be enabling and creating the future of servers and supercomputers around the globe.
Advanced Processor Lab (APL) is a lab that was recently launched and is committed to shaping the future of CPU processors and SoC architecture. APL started working on RISC-V processor research and its main mission is to design the RISC-V
micro-architecture and its advanced features. Advanced Material Lab (AML) is located in Boston. AML has mainly been focusing its efforts on material research and development, with a special emphasis on next-gen semiconductors and batteries.
Our engineers use computational simulation methodologies and high-throughput synthesis to explore and discover novel materials for these applications.
Meta Vision Lab (MVL) in Pasadena is focusing on developing next-gen optical interconnect architectures, circuits, and devices to support ultra-high data bandwidth requirements necessary for HPC and AI/ML applications. Our goal is to build
the interconnect with the size and power efficiency similar to that of electrical interconnects, but with the extended reach of traditional optical links. The research areas involved are system level modeling and simulation, high speed
circuit design, and ML-based optimization.
In addition to R&D, SAIT-America acts as the liaison between SAIT-HQ and the most innovative players in North America for our open innovation efforts. We strive to create strong research relationships with top U.S. based universities and
startups by embarking on a journey filled with creative and innovative collaborations. We look forward to partnering with you.
Address & Contact
-
Silicon Valley
-
665 Clyde Ave. Mountain View, CA 94043
-
+1-650-584-2141
-
-
Boston
-
10 Wilson Rd. Suite 150 Cambridge MA 02138
-
+1-650-336-3508
-
-
Pasadena
-
2N Lake Ave, Ste 230, Pasadena, CA 91101
-
+1-626-683-3465
-
Montreal Lab
- Deep Learning
- Machine Learning
Canada
Overview
SAIL Montreal (SAIT AI Lab Montreal) is an academic-style research lab (in close collaboration with Mila) whose mission is to push the boundaries of AI with SAIT and Mila through open fundamental research. Headed by Simon Lacoste-Julien, a professor in computer science at Université de Montréal and co-founding member of Mila, SAIL is located in Mila's corporate space at the heart of the Montreal AI ecosystem nearby Element AI, Borealis AI, FAIR, Microsoft Research and others where an open collaborative environment is encouraged.
Research Areas & Core Activities
Current research themes for the lab are: optimization and generalization in machine learning, with applications to structured data such as in natural language understanding or computer vision; interpretability of deep models, causal
modeling and explainable AI; neural network compression and graph neural networks.
Interaction with visitors from the AI and Software research center from SAIT (Samsung Advanced Institute of Technology) in Korea is encouraged.
Their research interests include deep learning, natural language processing, visual recognition, neural processor and autonomous driving, among others.
People
Geneviève Bernard
Office Manager
Code
Github
HuggingfacePublications
Generating and Imputing Tabular Data via Diffusion and Flow-based Gradient-Boosted Trees
Authors Alexia Jolicoeur-Martineau, Kilian Fatras, Tal Kachman
Conference AISTATS 2024
Fairness in Submodular Maximization over a Matroid Constraint
Authors Marwa El Halabi, Jakub Tarnawski, Ashkan Norouzi-Fard, Thuy-Duong Vuong
Conference AISTATS 2024
Adaptive Quasi-Newton and Anderson Acceleration Framework with Explicit Global (Accelerated) Convergence Rates
Authors Damien Scieur
Conference AISTATS 2024
Networks for Networks: Graph Neural Networks for Learning Equivariant Representations of Neural Networks
Authors Miltiadis Kofinas, David W. Zhang, Boris Knyazev, Yan Zhang, Yunlu Chen, Gertjan J. Burghouts, Efstratios Gavves, Cees G. M. Snoek
Conference ICLR 2024
How connectivity structure shapes rich and lazy learning in neural circuits
Authors Yuhan Helena Liu, Aristide Baratin, Jonathan Cornford, Stefan Mihalas, Eric Todd SheaBrown, Guillaume Lajoie
Conference ICLR 2024
Object centric architectures enable efficient causal representation learning
Authors Amin Mansouri, Jason Hartford, Yan Zhang, Yoshua Bengio
Conference ICLR 2024
Multi-View Causal Representation Learning with Partial Observability
Authors Dingling Yao, Danru Xu, Sebastien Lachapelle, Sara Magliacane, Perouz Taslakian, Georg Martius, Julius von Kügelgen, Francesco Locatello
Conference ICLR 2024
Using Representation Expressiveness and Learnability to Evaluate Self-Supervised Learning Methods
Authors Yuchen Liu, Zhen Liu, Aristide Baratin, Romain Laroche, Aaron Courville, Alessandro Sordoni
Conference TMLR (Nov-23)
Decision-Aware Actor-Critic with Function Approximation and Theoretical Guarantees
Authors Sharan Vaswani, Amirreza Kazemi, Reza Babanezhad Harikandeh, Nicolas Le Roux
Conference NeurIPS 2023
Additive Decoders for Latent Variables Identification and Cartesian-Product Extrapolation
Authors Sébastien Lachapelle, Divyat Mahajan, Ioannis Mitliagkas, Simon Lacoste-Julien
Conference NeurIPS 2023
Can We Scale Transformers to Predict Parameters of Diverse ImageNet Models?
Authors Boris Knyazev, Doha Hwang, Simon Lacoste-Julien
Conference ICML 2023
Difference of submodular minimization via DC programming
Authors Marwa El Halabi, George Orfanides, Tim Hoheisel
Conference ICML 2023
Fairness in Streaming Submodular Maximization over a Matroid Constraint
Authors Marwa El Halabi*, Ashkan Norouzi-Fard*, Jakab Tardos*, Jakub Tarnawski*, Federico Fusco*
Conference ICML 2023
Unlocking Slot Attention by Changing Optimal Transport Costs
Authors Yan Zhang*, David W Zhang*, Simon Lacoste-Julien, Gertjan J Burghouts, Cees GM Snoek
Conference ICML 2023
Equivariance with Learned Canonicalization Functions
Authors Sékou-Oumar Kaba*, Arnab Kumar Mondal*, Yan Zhang, Yoshua Bengio, Siamak Ravanbakhsh
Conference ICML 2023
CrossSplit: Mitigating Label Noise Memorization through Data Splitting
Authors Jihye Kim, Aristide Baratin, Yan Zhang, Simon Lacoste-Julien
Conference ICML 2023
Fast Online Node Labelling with Local and Global Consistency
Authors Baojian Zhou, Yifan Sun, Reza Babanezhad
Conference ICML 2023
Target-based Surrogates for Stochastic Optimization
Authors J. Wilder Lavington, Sharan Vaswani, Reza Babanezhad, Mark Schmidt, Nicolas Le Roux
Conference ICML 2023
Lazy vs hasty: linearization in deep networks impacts learning schedule based on example difficulty
Authors Thomas George, Guillaume Lajoie, Aristide Baratin
Journal TMLR (Dec-22)
Nonlinear acceleration of momentum and primal-dual algorithms
Authors Raghu Bollapragada, Damien Scieur, Alexandre d’Aspremont
Journal Mathematical Programming (Feb-22)
Acceleration in Optimization
Authors Alexandre d'Apresmont, Adrien Taylor, Damien Scieur
Journal Foundation and Trends© in Optimization (Dec-21)
SVRG meets AdaGrad: Painless Variance Reduction
Authors Benjamin Dubois-Taine, Sharan Vaswani, Reza Babanezhad, Mark Schmidt, Simon Lacoste-Julien
Journal Machine Learning Journal (Nov-22)
The Curse of Unrolling: Rate of Differentiating Through Optimization
Authors Damien Scieur, Quentin Bertrand, Gauthier Gidel and Fabian Pedregosa
Conference Neurips 2022
Data-Efficient Structured Pruning via Submodular Optimization
Authors Marwa El Halabi, Suraj Srinivas, Simon Lacoste-Julien
Conference Neurips 2022
Hyper-Representations as Generative Models: Sampling Unseen Neural NetworkWeights
Authors Konstantin Schürholt, Boris Knyazev, Xavier Giró-i-Nieto, Damian Borth
Conference Neurips 2022
MCVD: Masked Conditional Video Diffusion for Prediction
Authors Vikram Voleti, Alexia Jolicoeur-Martineau, Christopher Pal
Conference Neurips 2022
Model Zoos: A Dataset of Diverse Populations of Neural Network Models
Authors Konstantin Schürholt, Diyar Taskiran, Boris Knyazev, Xavier Giró-i-Nieto, Damian Borth
Conference Neurips 2022
Towards Painless Policy Optimization for Constrained MDPs
Authors Arushi Jain, Sharan Vaswani, Reza Babanezhad Harikandeh, Csaba Szepesvari, Doina Precup
Conference UAI 2022
Only tails matter: Average-Case Universality and Robustness in the Convex Regime
Authors Leonardo Cunha, Gauthier Gidel, Fabian Pedregosa, Courtney Paquette, Damien Scieur
Conference ICML 2022
Towards Noise-adaptive, Problem-adaptive (Accelerated) SGD
Authors Sharan Vaswani, Benjamin Dubois-Taine, Reza Babanezhad
Conference ICML 2022
Image Retrieval from Contextual Descriptions
Authors Benno Krojer, Vaibhav Adlakha, Vibhav Vineet, Yash Goyal, Edoardo Ponti, Siva Reddy
Conference ACL 2022
Super-Acceleration with Cyclical Step-sizes
Authors Baptiste Goujaud, Damien Scieur, Aymeric Dieuleveut, Adrien Taylor, Fabian Pedregosa
Conference AISTATS 2022
MULTISET-EQUIVARIANT SET PREDICTION WITH APPROXIMATE IMPLICIT DIFFERENTIATION
Authors Yan Zhang, David W Zhang, Simon Lacoste-Julien, Gertjan J. Burghouts, Cees G. M. Snoek
Conference ICLR 2022
On Evaluation Metrics for Graph Generative Models
Authors Rylee Thompson, Boris Knyazev, Elahe Ghalebi, Jungtaek Kim, Graham W. Taylor
Conference ICLR 2022
Infinite-Dimensional Optimization for Zero-Sum Games via Variational Transport
Authors Lewis Liu, Yufeng Zhang, Zhuoran Yang, Reza Babanezhad, Zhaoran Wang
Conference ICML 2021 (Jul-21)
Connecting Sphere Manifolds Hierarchically for Regularization
Authors Damien Scieur, Youngsung kim
Conference ICML 2021 (Jul-21)
Affine Invariant Analysis of Frank-Wolfe on Strongly Convex Sets
Authors Thomas Kerdreux, Lewis Liu, Simon Lacoste-Julien, Damien Scieur
Conference ICML 2021 (Jul-21)
Repurposing Pretrained Models for Robust Out-of-domain Few-Shot Learning
Authors Namyeong Kwon, Hwidong Na, Gabriel Huang, Simon Lacoste-Julien
Conference ICLR 2021 (Sep-21)
Average-case Acceleration for Bilinear Games and Normal Matrices
Authors Carles Domingo-Enrich, Fabian Pedregosa, Damien Scieur
Conference ICLR 2021 (Sep-21)
An Analysis of the Adaptation Speed of Causal Models
Authors Remi Le Priol, Reza Babanezhad, Yoshua Bengio, Simon Lacoste-Julien
Conference AISTATS 2021 (Apr-21)
Generalization of Quasi-Newton Methods: Application to Robust Symmetric Multisecant Updates
Authors Damien Scieur, Lewis Liu, Thomas Pumir, Nicolas Boumal
Conference AISTATS 2021 (Apr-21)
Extra-gradient with player sampling for provable fast convergence in n-player games
Authors Samy Jelassi, Carles Domingo Enrich, Damien Scieur, Arthur Mensch, Joan Bruna
Conference ICML 2020 (Jul-20)
Universal Average-Case Optimality of Polyak Momentum
Authors Damien Scieur, Fabian Pedregosa
Conference ICML 2020 (Jul-20)
Average-case Acceleration Through Spectral Density Estimation
Authors Fabian Pedregosa, Damien Scieur
Conference ICML 2020 (Jul-20)
Accelerating Smooth Games by Manipulating Spectral Shapes
Authors Waïss Azizian, Damien Scieur, Ioannis Mitliagkas, Simon Lacoste-Julien, Gauthier Gidel
Conference AISTATS 2020 (Aug-20)
Address & Contact
-
Montreal-
6666 St Urbain St, Suite 200, Montreal, QC H2S 3H1
-
g.bernard@samsung.com
-
Yokohama Lab
- Semiconductor Materials
- Display Materials
Japan
Overview
Samsung R&D Institute Japan (SRJ) was established in August 1992, and completed construction of its building in Yokohama in June 1997. SRJ was pivotal in accelerating research and development activities rooted in Japan. In January 2023, Samsung Device Solutions R&D Japan (DSRJ) was newly established for Samsung Electronics's DS division in order to focus a wider and higher level of research and development.
Research Areas & Core Activities
Research activity is being focused in the following areas; System LSI, Advanced Packaging, TCAD Simulator, Electronic Materials and Devices for next-gen semiconductor, display, and battery technologies.
In particular, SAIT AEML (Advanced Electronic Materials Lab) is focusing on developing cutting-edge semiconductor and display materials.
We believe that innovation comes from the creativity and insight stimulated through the interaction with academia.
Partnerships with leading universities and institutes are indispensable for DSRJ's research and development activities. DSRJ is actively conducting research partnerships with universities and research institutes in Japan.
Address & Contact
-
Yokohama-
2-7 Sugasawa-cho, Tsurumu-ku, Yokohama, Japan
-
contact.dsrj@samsung.com
-
Beijing Lab
- Computer Vision
Xian Lab
- SW & Algorithm
China
Overview
SAIT has two Labs in China, SAIT-Beijing and Samsung R&D Institute China Xi'an.
SAIT-Beijing was founded in June 2008 with the directive to "Set up a R&D base for future core-technology & China's leading technologies". The organization is comprised of an advanced research lab and open innovation for future
emerging tech. SAIT-Beijing is located in Chaoyang District, Beijing - the R&D center of China teeming with top universities, prestigious research institutions, renowned companies, and fast-evolving startups; providing an abundant
pool of talented researchers and collaboration opportunities for Samsung and SAIT.
Samsung R&D Institute China Xi'an (SRCX) was established in 2013.
Xi'an, situated at the heart of western China's R&D hub, boasts prestigious universities, national-level research institutions, and international enterprises. Capitalizing on Xi'an's and China's software talent pool, our institute
is deeply entrenched in the local cultural and industrial milieu, attracting a cohort of elite software engineers and researchers with both international perspectives and local expertise.
Research Areas & Core Activities
The SAIT-Beijing Lab in Beijing focuses on advanced research for next-generation technologies in computer vision, artificial intelligence, and AI for science. The lab's achievements include contributing core technologies towards Samsung's
flagship products. SAIT-Beijing Lab achieved rank 1 results on public benchmark datasets such as KITTI Stereo Matching (2021), ICCV SLAM Challenge (2023), and ETH3D SLAM (2023). The research results are published in world’s top conferences
including ICML, NeurIPS, CVPR, and ICCV.
SAIT-Beijing aims to achieve scientific innovations through exploratory research in order to advance Samsung's existing technologies and become one of the most influential research labs in China. We collaborate with China's top universities
and research institutes to work on world-leading innovative technologies.
SAIT-Beijing is actively enhancing its local R&D network to incubate differentiating innovative core technologies based on China specific scientific and technical strengths.
In parallel, SRCX aims to innovate the world's top fundamental technologies that will be utilized by various future business demands within Samsung. Utilizing Samsung's next-gen hardware, we are building AI/HPC computing platforms as well
as AI technologies such as human-centric vision perception, powerful generative AI platform, and efficient model lightweight technologies. SRCX constantly delivers research achievements in the form of international patents and top-tier
conference (CVPR, ICCV, AAAI, etc.) papers every year.
SRCX serves not only as a hub for technological innovation but also as a bridge deeply embedded in the Chinese market. With our advanced software R&D capabilities and a focus on global alignment, we collaborate closely with SAIT HQ to
ensure that our developments maintain global competitiveness.
Address & Contact
-
Beijing-
18F, Sun Palace Building, No.12-A, Taiyanggong Mid Road, Chaoyang District, Beijing, China 100028
-
+86-10-8443-9777
-
-
Xian-
B-1 Building, Xian Software Town Phase 2, No. 156 Tiangu 8th Rd, Hi-tech, Xian, China
-
+86-029-8874-1490
-
Bengaluru Lab
- System Hardware
India
Overview
Samsung started its Bengaluru center in 1996, primarily leveraging India's growing IT prowess. Since then, the center has grown to become one of the largest R&D center of Samsung outside Korea. Today, 4000+ strong workforce is
providing a wide variety of solutions to various main business units in Samsung.
The dedicated R&D group called SAIT-India was established in 2009 with an aim to primarily conduct R&D in future IT technology. Since 2020, SAIT-India has become an integral part of Samsung Semiconductor India Research (SSIR), a
dedicated DS organization in India, to synergize with the core strengths of SSIR.
SSIR is part of the global network of Samsung, providing component solutions featuring industry-leading technologies such as System LSI, Memory and Foundry. At SSIR, we offer our engineers a foundation to work on cutting edge
technologies such as Foundation IP Design, Serial Interfaces, Multimedia IPs, Mobile SoCs, Storage Solutions, 4G/5G Solutions, Neural Processors, AI/ML, and more.
Research Areas & Core Activities
SAIT-India group is actively involved in advanced topics such as Near Memory Computing, AI accelerators, and RISC-V based SoC. In addition, SAIT-India is developing algorithms for safe battery operation using physics based simulations by
working closely with the research groups in SAIT, Korea. SAIT-India continues to attract top talent around the country and currently comprises of 40+ engineers with many of our researchers having advanced degrees or crucial work experience
from abroad. We are on course to becoming one of the leading R&D establishments in India with performance measured by the number of patents and publications.
SAIT-India, through SSIR, continues to build strong relationships with top Indian institutions; co-working on advanced research topics. Our vision is to build on India's IT competency & fabless design capabilities in order to drive
innovation in multiple disciplines; thus contributing to Samsung's vision of creating new technologies and reaching every individual around the globe.
Address & Contact
-
Bengaluru-
Angkor West, Block-C01, Gopalan Urban Woods, Garudachar Palya, Mahadevapura, Bengaluru, Karnataka 560048
-
Kyiv Lab
- Software Implementation
Ukraine
Overview
Samsung R&D Institute Ukraine (SRUKR) was established in Ukraine in 2009.
SRUKR is strengthening their technological development and enhancing the competitiveness of Samsung products through key research in security, AI, and Computer Vision/Machine Learning focused areas.
Research Areas & Core Activities
SRUKR is composed of prominent professionals and is working on the study of intelligent security, computer vision, context-aware intelligent service, and etc.
As part of the industrial-educational cooperation and CSV activity, SRUKR actively cooperates with local universities and schools, while creating high-level education activities, and making investments for the future of the IT sphere within
Ukraine.
Address & Contact
-
Kyiv-
57 Hetmana Pavla Skoropadskoho street, Kyiv, Ukraine, 01032
-
+380-44-392-1736
-
Warsaw Lab
- Signal Processing Algorithm
Poland
Overview
SAIT created a R&D team in Warsaw, Poland in 2020 to tap into the unique skills and capabilities of Polish engineers. Since its inception, SAIT-Poland team is leading groundbreaking research in the field of camera and visual intelligence. Located in heart of Warsaw and close to university partners, its mission is to enable next-generation camera features and to introduce neural rendering to mobile products that will benefit millions of people all around the globe.
Research Areas & Core Activities
The SAIT-Poland team is focusing their research on signal processing challenges with AI and redefining the classical camera pipeline with next-generation AI-based camera pipeline for future mobile phones; neural rendering for mobile and AR devices that leverages advanced AI in order to optimize rendering processes, which ensures efficient performance on resource-constrained mobile hardware and broadening accessibility and usability in everyday applications; revolutionizing user experiences by providing highly realistic and interactive visual content that enhances immersion and engagement.
Address & Contact
-
Warsaw-
Samsung R&D Insitute Poland plac Europejski 1, 00-844 Warsaw
-
London Office
- Open Innovation
- R&D Strategy
United Kingdom
Overview
SAIT-UK was established to actively scout for future technologies throughout Europe since 2007. With the implementation of our open innovation and R&D strategies, we engage in exploration activities and find opportunities for collaborative future technology research with world-class research institutions, universities, and prominent start-ups around the continent.
Research Areas & Core Activities
SAIT-UK's open innovation activities and R&D strategies synergize to create innovative research programs within SAIT-HQ. SAIT-UK's engagement with R&D leaders in Europe allows SAIT-HQ to refine its R&D strategies and to seek out world-class
partners for research collaborations. SAIT-UK facilitates a range of collaborative opportunites including bilateral R&D project funding and Samsung's annual Global Research Outreach (GRO) funding program.
Our mission is to promote Europe's excellence in both fundamental and disruptive R&D within Samsung while identifying and developing relationships that will enable SAIT to create and deliver exceptional research.
Address & Contact
-
London-
Samsung R&D Institute UK, Communications House, South Street, Staines-upon-Thames, Surrey, TW18 4QE, United Kingdom
gro.europe@samsung.com
-