New business branch: AI department

Published in

Adhive.tv

7 min readMay 16, 2018

While developing the Platform, we have received lots of requests from potential customers for additional AI services, so we came to a conclusion that our AI may be helpful beyond the scope of the Platform.

Realizing that a technological startup should always be on the cutting edge of technology, we have decided to create our own AI department on the basis of existing solutions in order to ensure AdHive’s technological competition, as well as to increase the demand for AdHive services. All payments for such services will be executed in ADH.

Taking into account that most AI startup are being acquired by such huge corporations as Amazon and Google, AdHive will be one of a few independendent AI suppliers. We believe that such business diversification will raise brand awareness and strengthen AdHive’s positions.

The AI department will deal with the following areas:

Recognition of subjects on video and photos;
Recognition of speech;
Recognition of melodies;
Recognition of actions in videos;
Tools for creating neural networks and ML datasets;
Memory augmented neural networks (neural networks supplemented by memory);
Services for storing and distributing training data;
Frameworks for combining different machine learning algorithms into one ecosystem = computation framework.

What we have created so far:

AI Cluster

In order to provide an industrial solution for big data analysis of photo, audio and video content without losing data with the help of AI, AdHive has developed and created a software architecture called the AI Cluster.

The AI Cluster provides an opportunity for scaling the analysis ecosystem to a limitless number of servers.

Smart Sdk

The Smart is the lowest-level block of this system. It combines various algorithms for recognizing video and audio using convolution and recurrent neural networks. The Smart block consists of the main control program and modules executed as dynamically uploaded libraries.

Module = Algorithm

Each module has its own configuration file and is equipped with a standard interface. Thus, there is no need to make changes to the Smart itself to alter or create a new module.

The modules are added directly to the management file. Modules in Smart are algorithms for processing video, audio or text data. Each separate module is a dynamically loaded library with a standard interface.

Computation Framework v 1.0

The structure of artificial intelligence systems should be very different, depending on the tasks being solved. To simplify the creation of complex AI systems for prediction and classification, we have developed a software solution in the form of the Computation Framework.

The Framework performs 2 important functions:

1. Combination: combines various machine learning modules, as well as auxiliary modules into a single ecosystem.

2. Integration: provides integration of input and output data with business formats. Using the AdHive Computation Framework, AI developers can combine a variety of machine learning algorithms, such as neural networks, etc., thus creating their own constructs from various components.

TrainTool v 1.0

The TrainTool is an instrument for fast arrangement of datasets for ML modules.

TrainTool advantages:

Providing a high level of independence from a recognition services provider — teach on your own;
Creating additional data sets from input data;
Mining data automatically from the Internet via the URL;
Collecting data from user communities;
Optimised and convenient development;
Exchanging skills with other modules via cloud;
Training Neural Networks for Machine vision;
Training Neural Networks for Speech recognition;
Training classifiers for the Computation Framework;
Checking training results and statistics.

The TrainTool has a WEB interface to allow users to create and manipulate data sets.

What is coming this year:

AI Mobile SDK

The SDK for smartphones is being designed to provide all AI functions for mobile phone users. It is an embedded library for analysis of video, audio and photo materials.

It also can be used for the development of Android applications with elements of AI functionality.

Mobile Sdk version 1:

The Configuration of one of the Smart modules (specific algorithm for recognition of video, audio or photos) and its use in the application.

Mobile Sdk version 2:

The Configuration of multiple Smart modules, their setup and orchestration through the Computation Framework which is portable to Android.

The AI Mobile SDK also provides a mechanism for integrating with the Cloud Knowledge Base to provide access to training data.

Cloud Knowledge Base

The Cloud Knowledge Base is a software solution for learning data updating through network protocols and internet. Such Data consists of trained neural networks and classifiers along with configuration files.

The Cloud Knowledge Base provides learned data sets for users of AI modules and the SDK. These data sets are used in data analysis of video, audio and photos.

The user can also choose what data they need through a classifier of algorithms and themes for speed and storage optimization.

The user can also receive notifications (by subscription) for different learning segments. Users can do it on mobile devices or by customising integration of an industrial solution.

This concept gives an opportunity to separate AI-learning services from consumers and provide a mechanism of convenient broadcasting of learning data for mass consumers or create an easy deployment mechanism for companies.

AI depatrment Roadmap

AI Cluster:

2018 — Q2:

Multilayered sounds recognition;
Microservices ecosystem;
Monitoring tools.

2018 — Q3:

Administrative UI;
Installation tools.

2018 — Q4:

Multiple users AI mining feature.

Smart Sdk:

2018 — Q2:

Music recognition.

2018 — Q3:

Contour analysis + segmentation;
Multiple types of GPU support.

2018 — Q4:

Augment Reality Features;
Multiple types of GPU support.

Traintool:

2018 — Q2:

Formation of datasets for speech recognition;
Integration with neural networks, launching and viewing statistics;
Sampling of classifiers uploaded to csv and other formats;
Export of trained matrices;
Increased speed of training of up to 20 new subjects per day;

2018 — Q3:

Photo marking (squares, contours);
ML Classifiers and integration into the Computation Framework;
Configuration of layers and units of the Computation Framework;
Knowledge Cloud Configuration;
Collection and classification of data from users of the community;
Billing of training fees;
Temporal classifier training tools.

2018 — Q4:

Position classifier training tools;
Memory augmented Neural Nets training tools.

AI Mobile SDK:

2018 — Q2:

Test version of neural networks for recognizing photos in Android.

2018 — Q3:

Mobile Sdk for recognizing and locating objects on photos and videos (streams).

2018 — Q4:

Augment Reality Features;
Contour analysis and segmentation;
Computation Framework Integration.

Computation Framework:

2018 — Q3:

Temporal classifiers;
Action recognition.

2018 — Q4:

Positions classifiers;
Memory augmented Neural Nets development launch.

Knowledge Cloud:

2018 — Q3:

Implementation of Knowledge Cloud version 0.8;
Import of data from the Cloud via mobile devices and the server through the Knowledge Cloud;
Export of data to cluster servers.

2018 — Q4:

Versioning data sets;
Data sets classifier;
Payment system for data obtained from users.

Our goal

We are aiming at providing small and middle sized businesses with efficent AI instruments and make it more accessible and easily integrated, compared to modern expensive heavyweight AI solutions available on the market. The first actual consumer of such instruments is AdHive Platform itslef, but apart from it there is already a plenty of interest to our solutions from:

our partners
regular people who find us on the internet
other blockchain projects
major mobile device manufacturers

AdHive is going to hold and take part in meetups and conferences, where we will introduce our AI solutions, and explain to businesses how they can easily integrate and use them. In view of the above it is obvious that one of our main goals is to make our technologies user friendly so everyone will be able to configure SDK, get comfortable with Train Tool or install an AI Cluster in their office and connect to the global AI data mining network and Cloud Knwoledge Base.