Table 3 Per time-step regression

From: Closed-form continuous-time neural networks

Model

Mean Squared Error (MSE)

Time per epoch (min)

†ODE-RNN7

1.904 ± 0.061

0.79

†CT-RNN48

1.198 ± 0.004

0.91

†AugmentedLSTM44

1.065 ± 0.006

0.10

†CT-GRU49

1.172 ± 0.011

0.18

†RNN-Decay7

1.406 ± 0.005

0.16

†Bi-directional RNN53

1.071 ± 0.009

0.39

†GRU-D51

1.090 ± 0.034

0.11

†PhasedLSTM52

1.063 ± 0.010

0.25

†GRU-ODE7

1.051 ± 0.018

0.56

†CT-LSTM50

1.014 ± 0.014

0.31

†ODE-LSTM9

0.883 ± 0.014

0.29

coRNN57

3.241 ± 0.215

0.18

Lipschitz RNN58

1.781 ± 0.013

0.17

LTC1

0.662 ± 0.013

0.78

Transformer36

0.761 ± 0.032

0.80

Cf-S (current work)

0.948 ± 0.009

0.12

CfC-noGate (current work)

0.650 ± 0.008

0.21

CfC (current work)

0.643 ± 0.006

0.08

CfC-mmRNN (current work)

0.617 ± 0.006

0.34

  1. Modelling the physical dynamics of a walker agent in simulation. Numbers present mean ± s.d. (n = 5). The performance of the models marked by † is reported from ref. 9. Bold values indicate the lowest error and best time per epoch (min).
Back to article page