Skip to Main Content

DigiSpark - The Launchpad for NI Digital Innovators

Stage 1: Idea Submitted
Stage 1: Team Fully Formed
Stage 2 Kick-Off Bootcamp
Stage 2: Pitches
Gathering votes
Stage 3: Bootcamp & £10k
stage 3: Option Roundabout
Thanks for your vote
This voting milestone has now ended, and the idea has progressed to a private milestone
Return to idea list
Thanks for approving the idea
This milestone has now ended, and the idea has progressed to a private milestone
Return to idea list

GABI: a generalized artificial brain intelligence framework to enhance AI and XAI applications

Nothing - Team Fully Formed

Explainable artificial intelligence (XAI) and AI markets were valued at USD 39.9 and 3.55 billion in 2019, and expected to grow dramatically in the coming years, with revenue going mainly to developing software applications. These projections are intertwined as XAI was created to enhance our understanding of AI models and accelerate its adoption by human societies. As human beings trust what they can see or understand, even the most amazing intelligent machine must be understood before being adopted by humans.

While some machine learning algorithms (e.g., decision trees) may solve a problem and at the same time allow us to understand their “reasoning” (e.g., using the selected variables), only more complex algorithms like IBM’s Watson or Google Deepmind’s Alpha Go/Zero/Star/Fold have mastered the smartest experts in particular applications. How can Watson and Alpha do extremely well, and how could them help our societies? Answer 1: Using black-box algorithms such as Deep Artificial Neural Networks (DANNs) to learn a model. Answer 2: Using XAI to accelerate a trusted adoption. However, although XAI have “unlocked” these black boxes and uncovered bias in some cases that may hurt our trust, in fact the use of many current XAI implementations do not really enlighten our understanding [1].

GABI will merge AI, XAI and Computational Neuroscience (CN), to accelerate the development of AI application and enhance our understanding of learning processes. CN will allow us to develop simpler mathematical/computational theory of the brain functioning and information processing that can be implemented in large-scale DANNs algorithms. In practice, we will implement and incorporate artificial “electrodes” to monitor the activity correlated with our models’ dynamical states and learning processes. These electrodes will record information about artificial "eyes", other motor outputs, and other possible sensorial modalities, as well as about the model’s neural spiking activity at different multiscale resolution (e.g., synthetic EEG, ECoG, LFP, fMRI, etc.).

In an initial stage of development, our GABI models will be challenged with the same applications developed by OpenAI and other AI communities. GABI models could be offered in products for developing autonomous vehicles, diagnostic in medical applications, robotics, etc., to improve human beings’ living standards, including health and wellness. Apart from developing/selling GABI-based models as software tools and products, we could market several services to other startups and companies who may adopt our model, in order to provide solutions for applications in areas such as autonomous driving, voice recognition, etc.

Reference:

[1] Rudin, C., 2019. Nature Machine Intelligence, 1(5), pp.206-215

Which university are you from?

Ulster University

Please add your pitch presentation as an attachment to your submission below.

No answer chosen

Videos
edited on 3rd May 2021, 18:05 by Jose Sanchez
Jose Sanchez

Attach files

@

Kyle Beacham 2 months ago

Status label added: Nothing - Team Fully Formed

Reply 0

Kyle Beacham 1 month ago

The idea has been progressed to the next milestone.

Reply 0

Share