AI for all
This is how you post a benchmark update
Just a little preview of the machine learning test, and its level of accuracy.
The test will be ignoring the 1st load of the model for the moment, later on, it will be a performance metric too. Here is the speed of the #iPhoneXsMAx
Those test are now fully stable across all apple products, using iOS
So, on the most complicated #Machine Learning engines of #Humangnify, the DSP that is converting audio to Fourrier transform histogram is holding back the performance of the voice recognition. By restructuring the code a little, I think I can improve the speed and free it from the DSP gate. I really want to post all of the CoreML engines at one, I probably need one more day to get this done right.
Thanks for your patience.
#DLNPs are Deep Natural Language Processing Neuronal networks, we have written few of them, including some able to tell you the mood of people, this will allow you to get the temperature on people, based on your emails, recent audio interaction or social media account of this person.
Out of those #DNLP, we are going to extract statistics about the people you interact, like how many time a person changed his mind, or contradicted you. We have good accuracy, but obviously, we have to keep working to make progress. If you join out sponsors, we will keep you at the edge of what is possible. Some of those discovery are in Patent Pending phase.
So, one of the best workload we have in Humangnify is the DeepSpeech port, it was rewritten to allow the Apple DSP to do the signal processing, instead of doing the fourrier transform by hand … Unfortunatly, the speed on the new iPhone Xs Max does not seem to improve because the DSP is holding back the rest, so, I need to rebalance a little the algorithm, to allow more room for the Neuronal part. There is no point to benchmark something with it not being the bottle neck, so, I will be coding it differently tomorrow.
???? Using Text data input.
???? Using audio as input.
????️ Using picture or video as input.
ResNet is a short name for Residual Network. As the name of the network indicates, the new terminology that this network introduces is residual learning.
What is the need for Residual Learning?
Deep convolutional neural networks have led to a series of breakthroughs for image classification. Many other visual recognition tasks have also greatly benefited from very deep models. So, over the years there is a trend to go more deeper, to solve more complex tasks and to also increase /improve the classification/recognition accuracy. But, as we go deeper; the training of neural network becomes difficult and also the accuracy starts saturating and then degrades also. Residual Learning tries to solve both these problems.
What is Residual Learning?
In general, in a deep convolutional neural network, several layers are stacked and are trained to the task at hand. The network learns several low/mid/high level features at the end of its layers. In residual learning, instead of trying to learn some features, we try to learn some residual. Residual can be simply understood as subtraction of feature learned from input of that layer. ResNet does this using shortcut connections (directly connecting input of nth layer to some (n+x)th layer. It has proved that training this form of networks is easier than training simple deep convolutional neural networks and also the problem of degrading accuracy is resolved.
Extract age and gender of a face.
recognize cars model and brand in USA market.
Part of GoogleNet, this one recognize emotion out of a picture and provide an idea of the emotion in the picture.
This one classify how people feel, and what kind of emotions they try to express in a TEXT, using #DNLP.
Recognize places out of a picture, it is part of GoogleNet
This one give you a score of how appropriate this picture is regarding to nudify.
This one does recognize face emotions, using a different structure of network, I use it to validate the results of CNNEmotions.
A legend in picture machine learning, it does qualify objects in the picture, it is never too far from the best of art. This version comes from googleNet.
It is a #DNLP, tells you , out of your texts, the sentiment convey by a sentence.
Detect agreesivity toward the reader, in House developped.
The famous DeepSpeech, nodified and recoded to work with iPhone DSP and Apple convolution neuronal network engine. It recognizes pretty accurately what you say, while not buring your battery down fast. Tuned in house, and used to index what ever you want, your meeting, your day, your life …
Detect when the people speaking change to an other person, learn the new person on the fly, and then, generate a time table of different speakers in a recorded audio. It allows humangnify to provide you a perfectly indexed meeting.
This one is trivial, it detecting insults, it is home made, using a regular #DLNP, it takes a text and score it from 0 to 100, zero is polite text, 100 is seriously deep disgusting bad English language.
