Experiments

×

Error message

Deprecated function: The each() function is deprecated. This message will be suppressed on further calls in menu_set_active_trail() (line 2405 of /var/www/numahub.com/includes/menu.inc).

We have been experimenting with AI for a few years now. Some worked. Some failed. 

 

 

  • English Grammar App - NounShoun

    NounShoun is the world's first "Do it Yourself" english grammar App for Android built on Artificial Intelligence that helps anyone learning english to identify the Parts of Speech (Nouns, Verbs, Preposition etc) of any sentence. Once you enter the sentence it is evaluated and part of speech are found. The description of the various part of speech are also available.

    Visit NounShoun Website

  • Marine Research & Conservation

    We helped Snook & Gamefish foundation in collecting, analysing and predicting fish distribution on the basis of hobby fishing data. The data is being used by Florida Wildlife Research and other US Government agencies in implementing local fishing laws. The data is geo tagged to improve stock assessment and research initiatives.

    Visit Angler Action

     

  • Virtual Ruler - Fishing Application

    One of our Fishing and Marine research client - iAngler Tournament Systems, hosts Fishing tournaments worldwide. They were having problems with determining if anglers were reporting the correct length of the fish. Even when measured with a tape, they needed a better way to verify measurements. Using relative caliberation technique (and some clever mathematics) we were able to build a Virtual measurement system. The tournament managers loved it.

    Visit iAngler Tournament

     

  • Ping Pong

    This is another example of Reinforcement learning. The player on right (the red one) is an intelligent player that is trying to determine the hidden strategy by playing multiple games with the machine player  (the yellow one).  When the ball crosses the midline, it changes color. The red player should hit or miss a ball. The left side switches allow you to configure (and change at runtime) the policy when the player should hit and when to miss. After playing a few rounds, the player on right determines this policy. You can notice the learning (and learned) stage by monitoring the checkboxes on the right side of the page. 

    View Ping Pong Demo

  • Pole Balancing

    This demonstration shows how reinforcement learning helps in balancing a pole task. This cartpole example is a very common toy that is used to validate different learning examples. A virtual world with gravity, friction etc has been implemented. Click on see demo to view this example live.

    View Cart Pole Demo

  • Voice enabled Forms

    In this one we were evaluating the possibility to fill up forms by just talking to them. In the example below, a sentence spoken is checked to fill up a basic airline reservation form. The system had additional advantage that it could find nearby airports as well as so relative time searchers (2 days later).

    View Demonstration Video

  • BenchCamp

    This initiative was meant to be a platform for product comparison and evaluation. Using a Gamification approach, we were able to build a dynamic system which could do product recommendations on the basis of importance you give to a particular feature.

    Visit Benchcamp Website

  • NumaSpace

    This was our entry into the NASA Space Challenge. This app allowed you to convert your phone into a gesture controlled glove. Along with it, you could use voice commands to manage spaceship, lighting, speed as well as handle emergency situations. See the video for more demo here.

    View NumaSpace Video

  • Try Shry

    This was a fun app that could be used by you to check how specific patterns would look on you. You could start by providing a picture of yourself, choose the T-Shirt you are wearing and then just point the camera to some pattern and click. The pattern would replace the selected clothing giving you a virtual fitting room.

  • Skin Detection

    This algorithm uses a collection of Machine learning tools to identify skin areas in a picture. Using a clustering approach followed by a Neural Network, this approach achieves very high accuracy with low noise. This algorithm has great use for identifying the dress type.

     

     

     

  • Contextual Parsing - NLP

    We have been working on a contextual parsing algorithm, which can identify the deaper meaning in any text. This technology can help build question-answering system, intelligent search as well as chat bots. This patent pending technology has enabled us to build more meaningful representations.