Special Session at ESANN 2027: Where Kernels Meet Networks

[Frank-Michael Schleif]

Dear colleagues,

we would like to draw your attention to the following special session at ESANN 2027:

Where Kernels Meet Networks: Neural Tangent Kernels, Gaussian Processes and Beyond

Organizers:
Frank-Michael Schleif, Technical University of Applied Sciences Würzburg-Schweinfurt, Germany
Nils M. Kriege, University of Vienna, Austria
Johan Suykens, KU Leuven, Belgium

Kernel methods and neural networks have historically often been viewed as competing paradigms. A substantial body of recent work, however, has revealed increasingly close connections between them.

Infinitely wide neural networks can be described in terms of kernels and Gaussian processes, while Neural Tangent Kernels provide an infinite-dimensional kernel perspective on the training dynamics of wide networks. At the same time, current research increasingly moves beyond fixed limiting kernels towards learnable, structured and expressive kernel models, finite-width effects, alternative infinite-width limits, and scalable computational approaches.

The aim of this special session is to bring together these different perspectives and to encourage interaction between researchers working on kernel methods, Gaussian processes, neural-network theory, and related areas.

We welcome theoretical, methodological and applied contributions on topics including, but not limited to:

• infinite-width, arc-cosine, compositional and deep kernels;
• correspondences between neural networks, Gaussian processes and Neural Tangent Kernels;
• finite-width effects and deviations from the NTK regime;
• feature learning and mean-field limits beyond lazy training;
• deep kernel processes and learning kernel representations from data;
• heavy-tailed and stable infinite-width limits;
• deep and restricted kernel machines;
• graph kernels, graph neural networks and message-passing expressivity;
• random features, Nyström methods, low-rank approximations and sketching;
• indefinite, non-metric and structured kernels;
• Gaussian processes for deep models, uncertainty quantification and calibration;
• applications to molecular, graph, scientific and other structured data.

Paper submission deadline: 18 November 2026
ESANN 2027: 21–23 April 2027, Bruges, Belgium and online

Submissions to special sessions follow the same review procedure, format and submission rules as regular ESANN papers. Authors should indicate the corresponding special session when submitting.

Further information about the special session and submission procedure can be found on the ESANN website.

We would be very happy to see contributions addressing the increasingly rich interface between kernel learning, Gaussian processes and neural networks.

Best regards,

Frank-Michael Schleif
on behalf of the session organizers
Nils M. Kriege · Johan Suykens · Frank-Michael Schleif