Active Inference6 Stochastic Surprisal: An inferential measurement of Free Energy in Neural Networks Reference: https://arxiv.org/pdf/2302.05776.pdf Background and Objective This paper is to mimic how the human perceive the world and apply it to machine vision system. Human perceive the world, reconstruct the world based on our prior knowledge, select the appropriate action, and change the world for better observation. If the prediction error between the prior beliefs and real world, then "lear.. 2023. 9. 2. How to install Deep learning environment in ubuntu? 1. Move to /tmp folder in order to download and install temporary files. cd /tmp 2. Update Debian packages sudo apt update 3. Install curl (help sending and receiving http messages) and download sh file by using curl, or directly download from https://www.anaconda.com/products/individual sudo apt install -y curl sha256sum Anaconda3-2023.07-2-Linux-x86_64.sh # check checksum sh Anaconda3-2023.07-.. 2023. 8. 31. Mathematical Background Reference: https://www.youtube.com/watch?v=YVDAodLNRXs&list=PLZ9Bz1i4njbXfu3qHpZh5ulUM3hutcLBk&index=42 1. Multiplication rule 2. Marginalization rule 3. Chain rule 4. Expected value (= weighted average) 5. KL divergence Kullback-Leibler Divergence Indicates "How much two probability distributions diverge" An approximate distribution mimicking the true distribution comes first, and the true dist.. 2023. 8. 27. Introspective Learning: A Two-Stage Approach for Inference in Neural Networks Reference: arxiv.org/pdf/2209.08425.pdf Objective When the train and test dataset distributions are dissimilar, the predictions and the explanations are incorrect. This paper defines the network's reflection stage by implicitly extract features that answer all N introspective questions without explicitly training on said features. The output y from the reflection stage shall be more robust than .. 2023. 8. 27. [논문] Explainable Machine Learning for Hydrocarbon Prospect Risking 출처 : https://eartharxiv.org/repository/view/5561/ Explainable Machine Learning for Hydrocarbon Prospect Risking Add a Comment You must log in to post a comment. eartharxiv.org Abstract 2023. 7. 25. Probing the Purview of Neural Networks via Gradient Analysis Reference: https://arxiv.org/pdf/2304.02834.pdf Objective This study aims at analyzing the model capacity in terms of data-dependent point of view. That is, we should consider not only the model's architecture, but also how the model process and generalize the training data. For doing this, this study introduces purview, and explains the model capcity in two perspective: top-down purview and bot.. 2023. 7. 17. 이전 1 다음
