DEEPK 2024
The aim of DEEPK 2024 is to provide a multi-disciplinary forum where researchers of different communities can meet, to find new synergies between deep learning and kernel machines, both at the level of theory and applications.
Practical information:
Georganiseerd door:
Major progress and impact has been achieved through deep learning architectures with many exciting applications such as by generative models and transformers. At the same time it triggers new questions on the fundamental possibilities and limitations of the models, with respect to representations, scalability, learning and generalisation aspects. Through kernel-based methods often a deeper understanding and solid foundations have been obtained, complementary to the powerful and flexible deep learning architectures. Recent examples are understanding generalisation of over-parameterised models in the double descent phenomenon and conceiving attention mechanisms in transformers as kernel machines.
Topics include but are not limited to:
- Deep learning and generalisation
- Double descent phenomenon and over-parameterised models
- Transformers and asymmetric kernels
- Attention mechanisms, kernel singular value decomposition
- Learning with asymmetric kernels
- Duality and deep learning
- Regularisation schemes, normalisation
- Neural tangent kernel
- Deep learning and Gaussian processes
- Transformers, support vector machines and least squares support vector machines
- Autoencoders, neural networks and kernel methods
- Kernel methods in GANs, variational autoencoders, diffusion models, Generative Flow Networks
- Generative kernel machines
- Deep Kernel PCA, deep kernel machines, deep eigenvalues, deep eigenvectors
- Restricted Boltzmann machines, Restricted kernel machines, deep learning, energy based models
- Disentanglement and explainability
- Tensors, kernels and deep learning
- Convolutional kernels
- Sparsity, robustness, low-rank representations, compression
- Nystrom method, Nystromformer
- Efficient training methods
- Lagrange duality, Fenchel duality, estimation in Hilbert spaces, reproducing kernel Hilbert spaces, vector-valued reproducing kernel Hilbert spaces, Krein spaces, Banach spaces, RKHS and C*-algebra
- Applications
Invited speakers
(in progress - list to be completed)
- Mikhail Belkin (University of California San Diego)
- Volkan Cevher (EPFL)
- Florence d'Alche-Buc (Telecom Paris, Institut Polytechnique de Paris)
- Julien Mairal (INRIA)
- Massimiliano Pontil (IIT and University College London)
- Dingxuan Zhou (University of Sydney)
Programme
to be announced
Call for abstracts
The DEEPK 2024 program will include oral and poster sessions. Interested participants are cordially invited to submit an extended abstract (max. 2 pages) for their contribution. See the timeline for extended abstract submissions.
Please prepare your extended abstract submission in LaTeX, according to the following stylefile [LaTeX stylefile] and submit it in PDF format (max. 2 pages) on the following website [submission website to be opened] .
Schedule
- Deadline extended abstract submission:
Feb 8, 2024 - Notification of acceptance and presentation format (oral/poster):
Feb 22, 2024 - Deadline for registration:
Feb 29, 2024
- International Workshop DEEPK 2024:
March 7-8, 2024
Organising committee
- Johan Suykens, KU Leuven, Chair
- Alex Lambert, KU Leuven
- Panos Patrinos, KU Leuven
- Qinghua Tao, KU Leuven
- Francesco Tonin, KU Leuven
Other info
Please consult the DEEPK 2024 website https://www.esat.kuleuven.be/stadius/E/DEEPK2024 for info on the programme, registration, location and venue. The event is co-sponsored by ERC Advanced Grant E-DUALITY and KU Leuven.