VERSES Gave a Demo and Presentation this Morning - Atari Benchmark Challenge
Showed how Active Inference stacks up to Deep Learning in training and performance
If you missed it ➡️ this morning, Gabriel Rene, CEO of VERSES AI, gave a presentation and demo of how VERSES and Active Inference stacks up to deep learning in regard to historic benchmark Atari game challenges in the field of AI (benchmarks that matter to the AI community, but the real impact of their tech is far beyond these benchmarks).
🔸Their presentation demonstrated an Active Inference Agent taking a mere 12 minutes to acquire the training and achieve the performance level to beat a human. 👉🏻The same challenge that the best neural nets require 2 hours to complete.
Other notable differences:
▪️ Active Inference uses continual learning, which is not possible with deep learning (it’s not learning by replays of past game play and storing and buffering through all that training data, it’s getting better at predicting with every move in real time).
▪️ 1000% more sample efficiency
▪️ The entire learning process is explainable
▪️ AND the learning takes place on a laptop with a 16GB standard graphics card 🔥
NO GIANT MONOLITHIC DATABASE OR SUPERPOWERED GPUs REQUIRED.
🔴 Bottom line, VERSES used a well respected benchmark test in the world of AI to show that they can train Active Inference Agents to do what current state of the art deep learning AI systems can do in that same test, but they can do it in a fraction of the time and at a fraction of the energy cost. …
The bigger story is that these same Active Inference Agent achieve these benchmarks by doing things that just not possible for deep learning.
I’m going to be writing a full review of this presentation, so stay tuned…
#ai #ActiveInference #DeepLearning #futureofai #GenAI #IntelligentAgents #intelligentsystems #autonomousagents #VERSESAI
I am not a tech person but you relate very well, Thank you.
When you say they demonstrated the Atari 10k benchmark… you mean they showed this? If this is the case, then why is the Atari 10k challenge on Q3 of the roadmap if it’s achievable and publicly demonstrated now?
Love your work. Thank you. 😊