Compressing information in an end node using an autoencoder neural network

Abstract

In one embodiment, an apparatus includes: a sensor to sense real world information; a digitizer coupled to the sensor to digitize the real world information into digitized information; a signal processor coupled to the digitizer to process the digitized information into a spectrogram; a neural engine coupled to the signal processor, the neural engine comprising an autoencoder to compress the spectrogram into a compressed spectrogram; and a wireless circuit coupled to the neural engine to send the compressed spectrogram to a remote destination, to enable the remote destination to process the compressed spectrogram.

What is claimed is: 
     
         1 . At least one computer readable storage medium having stored thereon instructions, which if performed by a machine cause the machine to perform a method comprising:
 generating an autoencoder comprising an encoder and a decoder, and generating a classifier, wherein the encoder is to encode a spectrogram into a compressed spectrogram, the decoder is to decode the compressed spectrogram into a recovered spectrogram, and the classifier is to identify one or more properties of real world information from the recovered spectrogram;   calculating a first loss of the autoencoder and calculating a second loss of the classifier;   jointly training the autoencoder and the classifier based at least in part on the first loss and the second loss; and   storing the trained autoencoder and the trained classifier in a non-transitory storage medium.   
     
     
         2 . The at least one computer readable storage medium of  claim 1 , wherein the method further comprises jointly training the autoencoder and the classifier based on a weighted sum of the first loss and the second loss. 
     
     
         3 . The at least one computer readable storage medium of  claim 1 , wherein the method further comprises:
 calculating the first loss according to a correlation coefficient; and   calculating the second loss according to a binary cross-entropy.   
     
     
         4 . The at least one computer readable storage medium of  claim 1 , wherein the method further comprises sending a trained encoder portion of the autoencoder to one or more end node devices to enable the one or more end node devices to compress spectrograms using the trained encoder portion. 
     
     
         5 . The at least one computer readable storage medium of  claim 4 , wherein the method further comprises:
 receiving uncompressed spectrograms from the one or more end node devices; and   incrementally training one or more of the autoencoder or the classifier based at least in part on the uncompressed spectrograms.   
     
     
         6 . The at least one computer readable storage medium of  claim 5 , wherein the method further comprises sending an incrementally trained encoder portion of the autoencoder to the one or more end node devices. 
     
     
         7 . The at least one computer readable storage medium of  claim 1 , wherein the method further comprises generating the encoder asymmetrically from the decoder. 
     
     
         8 . A method comprising:
 generating, in a computer system, an autoencoder comprising an encoder and a decoder, and generating a classifier, wherein the encoder is to encode a spectrogram into a compressed spectrogram, the decoder is to decode the compressed spectrogram into a recovered spectrogram, and the classifier is to identify one or more properties of real world information from the recovered spectrogram;   calculating, in the computer system, a first loss of the autoencoder and calculating a second loss of the classifier;   jointly training, in the computer system, the autoencoder and the classifier based at least in part on the first loss and the second loss; and   storing the trained autoencoder and the trained classifier in a non-transitory storage medium.   
     
     
         9 . The method of  claim 8 , further comprising jointly training the autoencoder and the classifier based on a weighted sum of the first loss and the second loss. 
     
     
         10 . The method of  claim 8 , further comprising:
 calculating the first loss according to a correlation coefficient; and   calculating the second loss according to a binary cross-entropy.   
     
     
         11 . The method of  claim 8 , further comprising generating the encoder asymmetrically from the decoder. 
     
     
         12 . The method of  claim 8 , further comprising sending a trained encoder portion of the autoencoder to one or more end node devices. 
     
     
         13 . The method of  claim 12 , further sending a trained decoder portion of the autoencoder to at least some of the one or more end node devices. 
     
     
         14 . The method of  claim 12 , further comprising receiving compressed spectrograms from at least some of the one or more end node devices, the compressed spectrograms compressed using the trained encoder portion. 
     
     
         15 . The method of  claim 14 , further comprising receiving the compressed spectrograms based at least in part on the real world information sensed by the at least some of the one or more end node devices. 
     
     
         16 . The method of  claim 14 , further comprising incrementally training one or more of the autoencoder or the classifier, based at least in part on a quality metric of at least some of the compressed spectrograms received from the at least some of the one or more end node devices. 
     
     
         17 . The method of  claim 12 , further comprising:
 requesting one or more uncompressed spectrograms from at least some of the one or more end node devices; and   incrementally training at least one of the autoencoder or the classifier based at least in part on the one or more uncompressed spectrograms.   
     
     
         18 . A system comprising:
 at least one processor;   memory coupled to the at least one processor; and   one or more non-transitory storage media, wherein the one or more non-transitory storage media comprises instructions which if performed by the system cause the system to perform a method comprising:
 generating an autoencoder comprising an encoder and a decoder, and generating a classifier, wherein the encoder is to encode a spectrogram into a compressed spectrogram, the decoder is to decode the compressed spectrogram into a recovered spectrogram, and the classifier is to identify one or more properties of real world information from the recovered spectrogram; 
 calculating a first loss of the autoencoder and calculating a second loss of the classifier; 
 jointly training the autoencoder and the classifier based at least in part on the first loss and the second loss; and 
 storing the trained autoencoder and the trained classifier in the one or more non-transitory storage media. 
   
     
     
         19 . The system of  claim 18 , wherein the system comprises a remote cloud server, the remote cloud server to send the trained autoencoder to one or more end nodes coupled to the remote cloud server via a network. 
     
     
         20 . The system of  claim 19 , wherein the remote cloud server is to receive compressed spectrograms from at least some of the one or more end node devices and process the compressed spectrograms.