Identifikace limitací modelů hlubokého učení pro aproximaci odpovědí V1 neuronů
| Název práce v češtině: | Identifikace limitací modelů hlubokého učení pro aproximaci odpovědí V1 neuronů |
|---|---|
| Název v anglickém jazyce: | Identifying limitations of deep learning models for approximating responses of V1 neurons |
| Akademický rok vypsání: | 2024/2025 |
| Typ práce: | diplomová práce |
| Jazyk práce: | |
| Ústav: | Katedra softwaru a výuky informatiky (32-KSVI) |
| Vedoucí / školitel: | Mgr. Ján Antolík, Ph.D. |
| Řešitel: | skrytý - zadáno a potvrzeno stud. odd. |
| Datum přihlášení: | 29.04.2024 |
| Datum zadání: | 29.04.2024 |
| Datum potvrzení stud. oddělením: | 29.04.2024 |
| Zásady pro vypracování |
| A pivotal method for delineating the operational characteristics of sensory neurons involves system identification, wherein machine learning techniques are employed to analyze extensive neuronal recordings in response to sensory stimuli. Numerous studies have scrutinized the primary visual cortex (V1) of biological organisms, yielding models that can presently forecast approximately 70-80% of the explainable variance in V1 neurons. However, the extent to which deep learning models can accurately approximate various computational aspects remains uncertain.
This project aims to leverage a comprehensive, large-scale model of the primary visual cortex in cats to generate an extensive dataset. Subsequently, we will train a state-of-the-art Deep Neural Network (DNN) for predicting V1 neuron behavior. Through extensive analysis, we will deconstruct the model to ascertain precisely which fundamental properties of V1 neurons—such as orientation tuning and size tuning—it can reliably predict, and which it cannot. By undertaking this project, we seek not only to elucidate the computational representations embodied in state-of-the-art, biologically inspired V1 models but also to identify and address the existing limitations within current DNN architectures. This endeavor holds the potential to enhance the ability of such architectures to discern the intricate encoding mechanisms within the biological visual system. |
| Seznam odborné literatury |
| 1. Daniel A. Butts. Data-Driven Approaches to Understanding Visual Neuron Activity (2019). Annual Reviews of Vision Neuroscience, 5:20.1–20.27
2. Cadena, S. A., Denfield, G. H., Walker, E. Y., Gatys, L. A., Tolias, A. S., Bethge, M., & Ecker, A. S. (2017). Deep convolutional models improve predictions of macaque V1 responses to natural images. bioRxiv, 201764. https://doi.org/10.1101/201764 3. Jan Antolık, Cyril Monier1, Yves Fregnac, and Andrew P. Davison (2019). A comprehensive data-driven model of cat primary visual cortex. BiorXiv 4. Konstantin-Klemens Lurz, Mohammad Bashiri, Konstantin Willeke, Akshay K. Jagadish, Eric Wang, Edgar Y. Walker, Santiago A. Cadena, Taliah Muhammad, Erick Cobos, Andreas S. Tolias, Alexander S. Ecker, Fabian H. Sinz (2020) Generalization in data-driven models of primary visual cortex. bioRxiv |