Skip to content

A neural network coded from scratch, MSc project in collaboration with @RiccardoParente and @Sergio-Ltr (team name: ArbitraryValues)

Notifications You must be signed in to change notification settings

fexed/neuralnetwork-from-scratch

Repository files navigation

Best results

MONK1

Date Accuracy Hidden layers and units Learning rate Momentum Init. function Activ. function Notes
16/01/2022 100% [20] 0.1 0.8 Xavier Sigmoid
14/01/2022 71.63% [20] 0.1 0.8 Xavier Sigmoid
11/01/2022 74.80% [15, 15, 15] 0.01

MONK2

Date Accuracy Hidden layers and units Learning rate Momentum Init. function Activ. function Notes
16/01/2022 100% [10] 0.05 Norm. Xavier Sigmoid
14/01/2022 88.10% [20] 0.1 0.8 Xavier
11/01/2022 59.52% [15, 15, 15] 0.01

MONK3

Date Accuracy Hidden layers and units Learning rate Momentum Init. function Activ. function Notes
16/01/2022 94.21% [10] 0.01 Xavier Sigmoid
14/01/2022 90.67% [20] 0.1 0.8 Xavier
11/01/2022 74.17% [15, 15, 15] 0.01

CUP

Date Accuracy MEE Loss Hidden layers and units Learning rate Momentum Init. function Activ. function Notes
17/05/2022 0% 2.37% 0.07951 1.1123 (MEE) [23] 0.0025 Norm. Xavier Tanh L2(l = 1e-05)
13/05/2022 0.67% 2.37% 0.09735 0.6836 (MSE) [25,25] 0.001 Norm. Xavier Tanh
10/05/2022 0% 0.9846 (MSE) [30] 0.001 0.85 He Tanh Quick baseline test