Virtual functional Magnetic Resonance Imaging
Artificial Neural Networks take a lot of inspiration from biology although with much abstraction. Still even with such abstractions, it is not a trivial thing to understand exactly how anything but the simplest of perceptron operates. This makes ANNs, amongst others, a “black box” model because all we know is what comes in and what comes out with next to no knowledge about the internals.
While in recent decades there has been an increasing trend towards explainable AI (XAI), ANNs have been placed on the far end of desirable models because of that opacity. Alternatives encoding such as Genetic Programming1-3 may also produce convoluted solutions but their end results is, at least, potentially intelligible by a human. At the same time, ANNs are extremely powerful as illustrated by, relatively, recent results on playing games from Atari4 or competitive e-sports5 or extensive text generation6.
While not the first attempt at making ANNs more understandable(missing reference) I argue here that the use of Virtual functional Magnetic Resonance Imaging (VfMRI), has tangible benefits:
- It operates on irregular, emergent topologies
References
- Miller, J. F. & Thomson, P. Cartesian Genetic Programming. in Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) vol. 1802 121–132 (Springer Verlag, 2000).
- Poli, R. A Field Guide to Genetic Programming. Wyvern ([Lulu Press], lulu.com, [S.I.], 2008).
- Miller, J. F. Cartesian Genetic Programming. in Natural Computing Series vol. 43 17–34 (2011).
- Mnih, V. et al. Playing Atari with Deep Reinforcement Learning. (2013).
- OpenAI et al. Dota 2 with Large Scale Deep Reinforcement Learning. (2019).
- OpenAI et al. GPT-4 Technical Report. https://arxiv.org/abs/2303.08774v6 (2023).
| 1. | Miller, J. F. & Thomson, P. Cartesian Genetic Programming. in Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) vol. 1802 121–132 (Springer Verlag, 2000). |
| 2. | Poli, R. A Field Guide to Genetic Programming. Wyvern ([Lulu Press], lulu.com, [S.I.], 2008). |
| 3. | Miller, J. F. Cartesian Genetic Programming. in Natural Computing Series vol. 43 17–34 (2011). |
| 4. | Mnih, V. et al. Playing Atari with Deep Reinforcement Learning. (2013). |
| 5. | OpenAI et al. Dota 2 with Large Scale Deep Reinforcement Learning. (2019). |
| 6. | OpenAI et al. GPT-4 Technical Report. https://arxiv.org/abs/2303.08774v6 (2023). |