Realistic Handwriting Generation Using Recurrent Neural Networks and Long Short-Term Networks

Bodapati, Suraj; Reddy, S. Deepak Mohan; Sugamya, Katta

doi:10.1007/978-981-15-1480-7_55

“…Figure 7 shows the handwriting generation network architecture. The neural network contains a stack of three flip-flop layers with each layer getting input from the layer below followed by Mixed Density Network (MDN) 42 in the last layer. The MDN network is used to output parameters of the bivariate Gaussian model to estimate negative log-likelihood loss.…”

Section: Figure 7 Ffnn Architecture Of Handwriting Generationmentioning

confidence: 99%

The flip-flop neuron: a memory efficient alternative for solving challenging sequence processing and decision-making problems

Kumari

¹

,

Chandrasekaran

²

,

Chakravarthy

³

2023

Neural Comput & Applic

View full text Add to dashboard Cite

Sequential decision making tasks that require information integration over extended durations of time are challenging for several reasons including the problem of vanishing gradients, long training times and significant memory requirements. To this end we propose a neuron model fashioned after the JK flip-flops in digital systems. A flip-flop is a sequential device that can store state information of the previous history. We incorporate the JK flip-flop neuron into several deep network architectures and apply the networks to difficult sequence processing problems. The proposed architectures include flip-flop neural networks (FFNNs), bidirectional flip-flop neural networks (BiFFNNs), convolutional flip-flop neural networks (ConvFFNNs), and bidirectional convolutional flip-flop neural networks (BiConvFFNNs). Learning rules of proposed architectures have also been derived. We have considered the most popular benchmark sequential tasks like signal generation, sentiment analysis, handwriting generation, text generation, video frame prediction, lung volume prediction, and action recognition to evaluate the proposed networks. Finally, we compare the results of our networks with the results from analogous networks with Long Short-Term Memory (LSTM) neurons on the same sequential tasks. Our results show that the JK flip-flop networks outperform the LSTM networks significantly or marginally on all the tasks, with only half of the trainable parameters.

show abstract

The Flip-flop neuron – A memory efficient alternative for solving challenging sequence processing and decision making problems

Kumari

¹

,

Vigneswaran

²

,

Chakravarthy

³

2021

Preprint

View full text Add to dashboard Cite

Sequential decision making tasks that require information integration over extended durations of time are challenging for several reasons including the problem of vanishing gradients, long training times and significant memory requirements. To this end we propose a neuron model fashioned after the JK flip-flops in digital systems. A flip-flop is a sequential device that can store state information of the previous history. We incorporate the JK flip-flop neuron into several deep network architectures and apply the networks to difficult sequence processing problems. The proposed architectures include flip-flop neural networks (FFNNs), bidirectional flip-flop neural networks (BiFFNNs), convolutional flip-flop neural networks (ConvFFNNs), and bidirectional convolutional flip-flop neural networks (BiConvFFNNs). Learning rules of proposed architectures have also been derived. We have considered the most popular benchmark sequential tasks like signal generation, sentiment analysis, handwriting generation, text generation, video frame prediction, lung volume prediction, and action recognition to evaluate the proposed networks. Finally, we compare the results of our networks with the results from analogous networks with Long Short-Term Memory (LSTM) neurons on the same sequential tasks. Our results show that the JK flip-flop networks outperform the LSTM networks significantly or marginally on all the tasks, with only half of the trainable parameters.

show abstract

Handwriting Generation and Synthesis using Recurrent Neural Networks

Kumar

¹

,

Madaan

²

,

Kumar

³

et al. 2022

2022 13th International Conference on Computing Communication and Networking Technologies (ICCCNT)

0

View full text Add to dashboard Cite

Realistic Handwriting Generation Using Recurrent Neural Networks and Long Short-Term Networks

Cited by 5 publications

References 2 publications

The flip-flop neuron: a memory efficient alternative for solving challenging sequence processing and decision-making problems

The flip-flop neuron: a memory efficient alternative for solving challenging sequence processing and decision-making problems

The Flip-flop neuron – A memory efficient alternative for solving challenging sequence processing and decision making problems

Handwriting Generation and Synthesis using Recurrent Neural Networks

Contact Info

Product

Resources

About