Is it possible to quantize a projected NN?
I have trained a LSTM Model to filter some signals (approximately following this example: https://it.mathworks.com/help/signal/ug/denoise-eeg-signals-using-differentiable-signal-processing-layers.html). Then, since the aim is to implement the NN in an hardware, I reduced the number of learnables using the Projection method of compression. This gives me the possibility to reduce the learnables from 4.3k to 1.5k. Then, I improved the accuracy of the model fine-tuning the projected LSTM. Finally, I would like to quantize the model on 8 bits. When I use the quantize function in MATLAB I have the following error: "The class deep.internal.quantization.config.ConfigType has no Constant property or Static method named ‘InputProjector’."
Is it possible to quantize a projected NN?
Fairly new in this area, be kind :)I have trained a LSTM Model to filter some signals (approximately following this example: https://it.mathworks.com/help/signal/ug/denoise-eeg-signals-using-differentiable-signal-processing-layers.html). Then, since the aim is to implement the NN in an hardware, I reduced the number of learnables using the Projection method of compression. This gives me the possibility to reduce the learnables from 4.3k to 1.5k. Then, I improved the accuracy of the model fine-tuning the projected LSTM. Finally, I would like to quantize the model on 8 bits. When I use the quantize function in MATLAB I have the following error: "The class deep.internal.quantization.config.ConfigType has no Constant property or Static method named ‘InputProjector’."
Is it possible to quantize a projected NN?
Fairly new in this area, be kind 🙂 I have trained a LSTM Model to filter some signals (approximately following this example: https://it.mathworks.com/help/signal/ug/denoise-eeg-signals-using-differentiable-signal-processing-layers.html). Then, since the aim is to implement the NN in an hardware, I reduced the number of learnables using the Projection method of compression. This gives me the possibility to reduce the learnables from 4.3k to 1.5k. Then, I improved the accuracy of the model fine-tuning the projected LSTM. Finally, I would like to quantize the model on 8 bits. When I use the quantize function in MATLAB I have the following error: "The class deep.internal.quantization.config.ConfigType has no Constant property or Static method named ‘InputProjector’."
Is it possible to quantize a projected NN?
Fairly new in this area, be kind 🙂 lstm, quantizing networks, projection, neural networks, quantization, compression techniques MATLAB Answers — New Questions
​