AI Department Update: TrainTool Release 2.0
Artificial intelligence is conquering the world!
Don’t panic, there’s no Skynet invasion, so keep your clothes, boots and motorcycle. It’s a real, functioning technology that can be taught to do anything: drive a car, predict life expectancy, assemble a Rubik’s cube and even debate about healthcare or the nature of the universe.
AdHive is leading the way. We are actively developing the tools for training AIs and giving them the ability to perform various functions. It’s already implemented on our platform, so you can enjoy the advantages there. But it’s also an independent project, so it can be used for any of your needs.
Today, we will tell you about some of the new features we have developed and implemented in the constructs we are working on.
The new release of the TrainTool training service includes the following features:
Neural networks
Neural networks process video and audio data to recognize specified objects. However, modules need to be trained to look for items the user wants to find. The TrainTool can be used to achieve the desired effect and produce quick results. You can upload training samples into the TrainTool and conduct testing to understand whether the sample is sufficient.
This is quite a simple procedure:
• Launch a neural network and launch your own training process;
• Configure the neural networks dataset and tags;
• Observe training on the charts;
• View images and change test samples;
• Check the training logs;
• Test the training tag.
Sound recognition Training
The TrainTool can be used to train an AI module to recognize brand mentions in both video and audio files.
The learning process is similar to training a neural network to recognize imagery, as the user can create training samples and load them into the program to launch the training process and conduct subsequent testing.
The process works in several steps:
• Create a new tag for the sound;
• Upload your clip and use a sound or video editor;
• Сut the audio from the video and add text to the fragment you had cut;
• Upload the sound and convert its format;
• Create variations of the sound;
• Overlay the background sound;
• Check the result.
The TrainTool demonstration video will describe the entire process in exhaustive details. Watch it here:
AI Mobile SDK Update
We have created the first version of the AI Mobile SDK.
The SDK for smartphones is designed to provide mobile phone users all the functions of the AI. It is an embedded library for analysis of video, audio and photo materials. The program is a local neural network, which allows users to use any mobile gadgets to create applications able to identify desired objects in photos and videos.
The following layers have been implemented thus far:
• Convolutional with valid and same padding with identity and relu activations;
• Max pooling with valid padding support;
• Fully connected with support for activations of identity, relu and softmax.
A total of about 30 test cases containing about 100 assortments (comparisons of the data received with the expected results) have been carried out thus far. Simple matrices of small sizes of about 6x6 had been introduced at the input stage within the framework of the tests. All the tests were carried out successfully.
At the moment, we are working on the integration of training data with external AI frameworks. We are also developing standards for the training data so it can used in the AdHive ecosystem through the Knowledge Cloud. This will also include the ability to integrate the data with other AI service providers.
Computational Framework Update
As part of our development work, we have added the ability to configure the classifiers to identify tags from speech. This will allow identifying the mention of a brand name within any part of the video.
Starting from July, this function will be integrated into the AdHive ecosystem and the TrainTool to allow users to train the AI to recognize brand names within the influencers’ speech.
The AdHive team is working ceaselessly on perfecting the platform ecosystem and all of its components to make sure that its competitive advantage remains unattainable. The advanced technological solutions we are implementing will ensure the further growth and development of the opportunities and services the AdHive project will be able to offer to all advertisers and influencers willing to join the growing ranks of its participants.
