Dear mentors, My name is Thomas Katraouras, and I am a final-year undergraduate student at the Department of Electrical and Computer Engineering at the University of Thessaly in Volos. I am reaching out to express my interest in the project *Identifying transition points in the ZGB model using convolutional neural networks (CNNs)* and ask some questions related to it. I have a strong background in AI/ML and computer vision. I have worked on a variety of projects, including object detection, localization and mapping, autonomous systems and natural language processing. I have also worked on the development and optimization of deep learning models. In addition to hands-on project development, I have research experience in deep learning, mainly focusing on neural network compression and financial forecasting. I’m currently working on my proposal. I have read the referenced papers and examined the Apothesis code to familiarize myself with the ZGB model (both with diffusion and diffusionless). I also ran the three given tests (compile and run Apothesis, read SurfaceSpecies files in Python, call Apothesis from Python and read the latest Surface generated). I have gathered the following questions: 1. Is TensorFlow a strict requirement for building the CNN, or can I also consider PyTorch? I’m comfortable with both frameworks, but I might prefer to use whichever turns out to be the best fit for this project, if it’s acceptable to you. 2. I would like a clarification on what “equilibrium” refers to in this context. I have seen it used to refer specifically to the non-poisonous active state but also to the CO-poisoned and O-poisoned states. Do we treat “equilibrium” as only the active (non-poisonous) state, or are all three steady states (active, CO-poisoned, O-poisoned) included in the definition of “equilibrium”? 3. What lattice sizes (or range of sizes) do you expect to be run in Apothesis and which other parameters (eg. adsorption rates, diffusion rate, temperature, pressure, etc.) should we vary to generate enough and diverse data? I’m trying to get a sense of the overall computational cost, both for running Apothesis, and training the CNN on a potentially large dataset. 4. Do we care about states at times before the system reaches its steady state? If so, an LSTM or a combination of CNN and LSTM might be more suitable to capture time‐dependent patterns. Thank you in advance for the clarifications and I’m looking forward to your response. Best regards, Thomas Katraouras
---- Λαμβάνετε αυτό το μήνυμα απο την λίστα: Λίστα αλληλογραφίας και συζητήσεων που απευθύνεται σε φοιτητές developers \& mentors έργων του Google Summer of Code - A discussion list for student developers and mentors of Google Summer of Code projects., https://lists.ellak.gr/gsoc-developers/listinfo.html Μπορείτε να απεγγραφείτε από τη λίστα στέλνοντας κενό μήνυμα ηλ. ταχυδρομείου στη διεύθυνση <gsoc-developers+unsubscribe [ at ] ellak [ dot ] gr>.