Persagen.com 🇨🇦 🏳️⚧️ ☮🏳️🌈 on Nostr: Kolmogorov-Arnold Networks: More Interpretable Neural Network ...
Kolmogorov-Arnold Networks: More Interpretable Neural Network
https://arxiv.org/abs/2404.19756
https://news.ycombinator.com/item?id=41162676
https://spectrum.ieee.org/kan-neural-network
* MLP have fixed activation functions on nodes (neurons)
* KAN have learnable activation functions on edges (weights)
* KAN have no linear weights at all: ev. wt parameter replaced by a univariate function
* much smaller KAN can achieve comparable/better accuracy than much larger MLP
#ML #MLarchitectures #NeuralNetworks #KAN #KolmogorovArnold #MLP #perceptrons
https://arxiv.org/abs/2404.19756
https://news.ycombinator.com/item?id=41162676
https://spectrum.ieee.org/kan-neural-network
* MLP have fixed activation functions on nodes (neurons)
* KAN have learnable activation functions on edges (weights)
* KAN have no linear weights at all: ev. wt parameter replaced by a univariate function
* much smaller KAN can achieve comparable/better accuracy than much larger MLP
#ML #MLarchitectures #NeuralNetworks #KAN #KolmogorovArnold #MLP #perceptrons