The Best ICS Publications in 2024

Competition - Awarded Publications

Evaluation Committee Members

  • prof. RNDr. Jaromír Antoch, CSc.
  • prof. Mgr. Zdeněk Dvořák, Ph.D.
  • doc. RNDr. Iveta Mrázová, CSc.
  • prof. Ing. Mirko Navara, DrSc.
  • Carles Noguera, Ph.D.
  • doc. RNDr. Petr Pišoft, Ph.D.
  • doc. Mgr. Robert Šámal, Ph.D.

A. BEST THEORETICAL RESULT:

  • Milan Paluš, Martina Chvosteková, Pouya Manshour Causes of extreme events revealed by Rényi information transfer.
    The paper studies a novel way to interpret relations between various parts of a chaotic system. The fact that information theory can help to study the development of weather is fascinating. It seems that, at least in some situations, it behaves better than the previously used methods, like Granger causality. The paper is multidisciplinary, but the mathematical/statistical part is well balanced with the climatological part.
  • Jiří Šíma, Petra Vidnerová, Mrázek, V. Energy Complexity of Convolutional Neural Networks.; Jiří Šíma, Cabessa, J., Petra Vidnerová On energy complexity of fully-connected layers.
    While the prominence of deep neural networks in modern AI solutions has been much acclaimed, their ever-expanding size considerably increases the costs for their training and energy needed to run them. Both aspects are crucial across large-scale LLM-like and small-scale edge-like applications. To allow for hardware-independent efficacy assessment, Šíma et al. propose a rigorous theoretical framework of energy complexity. The introduced concept comprises two types of memory – DRAM and Buffer – and evaluates the overall energy costs through the terms for computation energy and data energy spent. For the outlined energy complexity model, the authors establish its asymptotically optimal bounds for fully connected layers and extend it to convolutional neural networks in the other paper. Theoretically justified results are experimentally validated. It appears to be a potentially important theoretical contribution to the development of state-of-the-art machine learning technologies. Both papers were published in reputable journals and were already cited.

B. BEST APPLICATION RESULT:

  • Pavel Šanda, Jaroslav Hlinka, Van der Berg, M., Škoch, A., Bazhenov, M., Keliris, G. A., Krishnan, G. P. Cholineric modulation support dynamic switching of resting state networks through selective DMN suppression.
    The study combines chemogenetic data from rats with a conductance-based network model of the rat and human connectomes to demonstrate that elevating acetylcholine from the basal forebrain sharply dampens Default Mode Network (DMN) activity while sparing sensory networks. This mechanistic link—from ACh-driven reductions in K+-leak and excitatory synaptic conductances to the whole-brain fMRI signature—explains how cholinergic neuromodulation flips the brain from internally focused rest to task-oriented states and provides a quantitative framework for exploring attention deficits and cholinergic decline in disease. This is a highly compelling study that integrates experimental data with a computational model of the human brain. Its findings have substantial potential to advance therapeutic strategies for cognitive disorders.
  • Dudášová, J., Zdeněk Valenta, Sachs, J. R. Improving presicion of vaccine efficacy evaluation using immune correlate data in time-to-event models.
    Developing efficient and safe vaccines is extraordinarily costly and requires extensive long-term testing. On the other hand, the exact role of population demography and its impact on vaccine efficacy remains questionable in many settings. To better comprehend the vaccination efficacy in demographic subgroups, Dudášová, Valenta, and Sachs extended their previous approach based on logistic regression by utilising immunogenicity data (instead of vaccination status) as a predictor in time-to-event models. Immunogenicity concerns the ability of a substance to provoke an immune response. The newly developed statistical method harnessing immunogenicity data demonstrates more accurate efficacy estimates with narrower confidence intervals for the demography subgroups involved in the two performed clinical trials (for zoster and dengue vaccines). The article addresses a pertinent practical issue in an exceptionally rigorous way and was published in a prestigious journal.

C. BEST RESULT OF THE YOUNG AUTHORS:

  • Gabriela Kadlecová, Lukasik, J., Pilát, M., Petra Vidnerová, Safari, M., Roman Neruda, Hutter, F. Surprisingly Strong Performance Prediction with Neural Graph Features.
    Recent advances in AI and ML demand the leveraging of increasingly more complex neural network architectures that should be fast to train, easy to interpret, and as accurate as possible. The ideal network architecture clearly depends on the task at hand. Recently, neural architecture search (NAS) techniques have attracted broad awareness due to an automatic exploration of viable network models. Various performance prediction characteristics have been investigated to ameliorate the high computational costs necessary to train the probed networks, among others, the zero-cost proxies. Motivated by their limits, Kadlecová et al. introduced new innovative neural graph features (GRAF) that allow for higher stability and interpretability of the found results. In addition to being straightforward and fast to evaluate, they outperform most other predictors across multiple benchmarks, particularly when combined with other zero-cost proxies. Scientists from multiple international institutions collaborated on the research. The paper was presented at the prestigious ICML conference in July 2024; since then, it has been cited twice. Its first author, Ms. Kadlecová, is a young author pursuing her PhD.

Evaluated Publications