Dan McAteer (@daniel_mac8)

GPT-5.4에 대해 '모델이 상태(state)를 지속(persist)할 수 있다'는 루머가 있으며, Jeff Dean이 Latent Space 팟캐스트에서 이와 관련된 내용을 언급해 AI 연구실들이 상태-공간 모델(State-Space Models) 통합을 검토하거나 발견했을 가능성을 제기함. 향후 모델 아키텍처 변화(상태 지속성 통합)에 관한 암시성 보고.

https://x.com/daniel_mac8/status/2028849150368862416

#gpt5.4 #statespace #llm #research #openai

@axoaxonic @adredish Fully agree 👍 Horner's framework really begs for a formal dynamical model: defining trajectories, #attractors, and #manifolds within that 3D space. Something that could turn his conceptual #StateSpace into a genuine #computational theory of #memory dynamics.

I didn’t know Redish's book ("Beyond the Cognitive Map") before your comment! Sounds highly relevant and I’ll definitely put it on my reading list 👌

Quick wrap-up of the year https://statntheorecol.github.io/posts/news-2025/ A new preprint on #statespace models #identifiability https://arxiv.org/abs/2508.08714 (using a spectral perspective) and another on #occupancy models for heterogeneous data https://arxiv.org/abs/2510.08151

🧠 New paper by Aidan J. Horner (2025, Trends in Cognitive Sciences) introduces a 3D neural #StateSpace for #episodic memories. It replaces linear #SystemsConsolidation models with a dynamic framework where #hippocampal, #neocortical, and episodic specificity dimensions evolve independently and non-linearly, allowing memories to shift, reverse, or re-engage hippocampal circuits.

🌍 https://www.cell.com/trends/cognitive-sciences/fulltext/S1364-6613(25)00284-0

#Neuroscience #CognitiveScience #Hippocampus #CogSci #compneuro #memory

Today I'll get myself to read the S4 (Structured State-Space Sequence model) and the Hyena papers.
I like Transformers generality since they are graph models, and I don't like how much this is overlooked even while applying these neural networks to both language sequences and images, for example.
So I'm quite interested in the above alternatives, not so much for their supposed ability to extend the length of input sequences processed for the same computing costs, but rather as *models*.

I would like to apply all those models (or the ones I might find arguably better suited) to the study of EEG/MEG/fMRI signals.

Let's see
#neuroscience #NeuroAI #deeplearning #transformers #statespace #s4 #hyena #TimeSeries

@jt252

If you mean the kind of #Nothing that #Physicists can talk about (some kind of #HilbertSpace or #StateSpace of possibilities where the #LawsOfPhysics can operate) then yes.

#StuartKauffman - Is #Emergence #Fundamental?

https://www.youtube.com/watch?v=GVL2Y5z2jLU&ab_channel=CloserToTruth

#Philosophy #PhilosophyOfScience #Biology #PhilosophyOfBiology #Darwin #Evolution #PhaseTransition #PhaseSpace #StateSpace #PhaseTransitions #Biosphere #Emergence #Reductionism #Emergentism #Ergodicity #Order #Organisation #Consciousness #CloserToTruth #RobertKuhn

Ball Balancing Wheel Puts a Spin on Inverted Pendulums

If you march sufficiently deep into the wilderness of control theory, you'll no doubt encounter the inverted pendulum problem. These balancing acts have emerged with a number of variants over the years, but just because it's been done before doesn't mean there's no space for something new. Here, [David Gonzalez], has taken this classic problem and given it an original own spin-literally-where the balancing act is now a ball balanced precariously upon a spinning wheel. (Video, embedded below.) Mix in a little computer vision for sensing, a dash of brushless motor control, a bit of math, and you have yourself a closed-loop system that's bound to turn a few heads.

[David's] implementation is a healthy mix of classic control theory with some modern electronics. From the theory bucket, there's a state-space controller to drive both the angle and angular velocity of the ball to zero. The "state" is a combination of four terms: the ball angle, the ball's angular velocity, the wheel angle, and the wheel's angular velocity. [David] weights each of these terms and sums them together to create an input value to adjust the motor velocity driving the wheel and balance the ball.

From the electronics bin, [David] opted for an ESP32 running Arduino, the custom Janus Brushless Motor Controller running SimpleFOC, and a Maix Bit Microcontroller with an added camera running MicroPython to compute the ball angle. Finally, if you're curious to dig into the source code, [David] has kindly posted the firmware on Github.

We love seeing folks mix a bit of control theory into an amalgamation of familiar electronics. And as both precision sensors and motor controllers continue to improve, we're excited to see how the landscape of projects changes yet again. Hungry for more folks closing the loop on unstable systems? Look no further than [UFactory's] ball balancing robot and [Gear Down for What's] two wheeled speedster.

#robotshacks #ballbalancing #bldccontroller #brushlessmotor #controltheory #statespace