What I would like to see is an integration of discrete knowledge bases (e.g. triplet subject-relationship-object associations) with deep neural networks. Hopfield networks passed my mind earlier today as a possible base architecture for storing such knowledge bases (which is surprisingly the first video here and a memorable clip at that). Classical neural nets can provide the capacity to identify/classify concepts from instance properties (or actions from a limited set of context input etc), but we appear to require conceptual and episodic memory than can be continually grown and drawn upon by the computational kernel. This is the major limitation of current generation GPT/language modelling. The agent should be able to deal with an arbitrarily sized knowledgebase. Ideally the network architectures should be similar to enable integration (e.g. binary weights to represent discrete memories). The addition/subtraction of the episodic/concept memories from the neural knowledge base should not affect the learnt/learnable processing of its memory in achieving its goal (ie providing effective zeroshot). There might be a way of structuring the lower or intermediate layers of the deep network to store these (unlimited set of) entities/associations.
Hey man, thanks for the video :D I wanna ask you, do you know Jesus? He's offering eternal life to all who repent and believe on Him. I'm not here to force you to believe, but we will all stand before Him one day with an account to give of our lives. All who call upon His Name shall be saved❤
Hey. Nice idea to have short clips! But after watching this I didn't quite understand how the Hopfield Networks work.
thanks for the feedback, will try to make the clips either better or shorter ;)
First, hahaha. 👏
haha, welcome :D
welcome letitia :)
I did not know that you made the clips channel
Welcome :)
Oh, that makes sense. Thanks for this clip!
What I would like to see is an integration of discrete knowledge bases (e.g. triplet subject-relationship-object associations) with deep neural networks. Hopfield networks passed my mind earlier today as a possible base architecture for storing such knowledge bases (which is surprisingly the first video here and a memorable clip at that). Classical neural nets can provide the capacity to identify/classify concepts from instance properties (or actions from a limited set of context input etc), but we appear to require conceptual and episodic memory than can be continually grown and drawn upon by the computational kernel. This is the major limitation of current generation GPT/language modelling. The agent should be able to deal with an arbitrarily sized knowledgebase. Ideally the network architectures should be similar to enable integration (e.g. binary weights to represent discrete memories). The addition/subtraction of the episodic/concept memories from the neural knowledge base should not affect the learnt/learnable processing of its memory in achieving its goal (ie providing effective zeroshot). There might be a way of structuring the lower or intermediate layers of the deep network to store these (unlimited set of) entities/associations.
I agree. very nice idea. I think we're still far away from that, though
Neural Turing machines.
I want to like the video but it is at 69 likes.
Hard call :D
Seems kinda of a FSM
Hi! This is a great idea, I am subscribed to your main channel but sometimes I don't have the time to see the full videos! Good luck and thank you!
thanks for subscribing
in the thumbnail you look like travis bickle from taxi driver 😂
Also wtf I cannot understand these comments 😂
So like a Boltzmann machine which is Energy Based Model?
love u r explanation...good job
12th
yay!
danke dir. Freitag examen ueber neural networks
Haha love the new series
thanks for joining :)
love the pentagram
Hey man, thanks for the video :D I wanna ask you, do you know Jesus? He's offering eternal life to all who repent and believe on Him. I'm not here to force you to believe, but we will all stand before Him one day with an account to give of our lives. All who call upon His Name shall be saved❤