Sakana AI’s Post

DiffUCO: A Diffusion Model Framework for Unsupervised Neural Combinatorial Optimization Quick read: https://lnkd.in/gEDWwdc8 Paper: https://lnkd.in/gbQJNvAN Code: https://lnkd.in/gsW599Yw

Nisal presenting our explainable genAI work at #AAAI2024.

📢 #HighlyCited 1.  A Study on Detection of Malicious Behavior Based on Host Process Data Using Machine Learning https://lnkd.in/gEMNiEZA 2. Prediction and Optimization of Matte Grade in ISA Furnace Based on GA-BP Neural Network https://lnkd.in/gKY-sUX3 3. Alzheimer’s Dementia Speech (Audio vs. Text): Multi-Modal Machine Learning at High vs. Low Resolution https://lnkd.in/gb_JQD8T

Learn the basic concepts, implementations, and applications of graph neural networks in this self-paced course. Build fundamental knowledge in hands-on interactive activities to get started using GNN as a graph analysis tool. Enroll now: https://nvda.ws/3HjyPg6

IGNN-Solver: A Novel Graph Neural Solver for Implicit Graph Neural Networks The most serious challenge regarding IGNNs relates to slow inference speed and scalability. While these networks are effective at capturing long-range dependencies in graphs and addressing over-smoothing issues, they require computationally expensiv... https://lnkd.in/eMPbVkqb #AI #ML #Automation

Explore how to construct a simple Rust-based neural network from scratch. Start building, implement core components, and train on a dataset today. Read the article here - https://lnkd.in/gUm-4JKG #Rust #NeuralNetwork #DEVCommunity

🌟 Exciting new research alert! Check out our latest blog post on how message passing enhances collaborative filtering methods. This in-depth article delves into the impact of message passing on recommender systems, focusing on its benefits for graph neural networks and real-world applications. Dive into our curated ablation studies and theoretical analyses to discover groundbreaking insights on the relationship between message passing and CF performance. Plus, don't miss our innovative Test-time Aggregation for CF framework, TAG-CF, designed to optimize graph knowledge without the computational overheads. Click here to read the full post: https://bit.ly/4aAC5iQ #collaborativefiltering #messsagepassing #recommendersystems #graphneuralnetworks #TAGCF

Check out the newly published blog post on "The fast committor machine: Interpretable prediction with kernels" on arXiv. The paper introduces a fast and interpretable method for approximating the committor, called the "fast committor machine" (FCM). This innovative approach utilizes simulated trajectory data to train a kernel model, identifying low-dimensional subspaces that optimally describe transitions from set A to set B. The FCM offers efficient training with linear scalability, outperforming neural network models in accuracy and speed. Dive into the details here: https://bit.ly/3yrBjGK

Good to be back after a much-needed break! During my time off, I immersed myself in some incredible technical material. I especially enjoyed reading Generating Sound and Organizing Time by Gregory Taylor and Graham Wakefield—highly recommend it to anyone looking to delve deeper into gen~ in Max. It does a fantastic job of demystifying computer music techniques and low-level signal processing. I also spent time familiarising myself with the FluCoMa Toolkit. This patch is a result of all the knowledge I absorbed during this period. In this iteration, I've built a network that uses the MLPRegressor in FluCoMa to control a granular synthesis module written in gen~. The MLPRegressor is a neural network that performs regression, i.e., it takes n-dimensional input data and maps it to n-dimensional output data. In this iteration, MLPRegressor takes 2-dimensional input data from an XY pad and maps it to 14 parameters of the granular synthesis module. On a tangential note, I'm excited to share that I will be facilitating two workshops on utilising neural networks and machine learning with Max/MSP. Stay tuned for more details! #MachineLearning #NeuralNetworks #MaxMSP #MusicTechnology #AudioProgramming #SoundDesign #DigitalSignalProcessing #ComputerMusic

Preprint 🚨: we propose a transformer pointer neural network that significantly improves the computational performance of the Column Generation approach for parallel machine scheduling. The proposed NN can solve different problem sizes as well as instances from different probability distribution than what it was trained on! https://lnkd.in/e2Qrrmcg