I just wish Cannon had called it homeodynamics, which appears like a better analogy for what’s happening. Better yet, William Butts (not a Bart Simpson reference!) use of the term homeorhesis better describes a system that returns to the same trajectory in phase space.
Whoa, this is very interesting! Is homeorhesis generally the result of some type of feedback in the system? Apologies, I'm not really sure how to formalize this question.
The term "homeorhesis" was coined by C.H. Waddington, it refers to control mechanisms used by biological organisms to remain close to a viable trajectory of development and growth throughout their lifetimes.
From Waddington's _The Strategy of the Genes_ (1957):
"We spoke earlier of trajectories in the phase space as converging on one or another definite end-point. As a matter of fact, if a process of embryonic development is disturbed, it usually returns to normality some time before reaching the adult condition. Its trajectory, that is to say, converges not merely to the normal end state, but to some earlier point on the path leading towards the steady state. This is well symbolised by the epigenetic landscape. If a ball, running down one of the valleys, were pushed partway up the hillside, it might well reach the valley bottom again before the slope of the valley flattens out as it reaches the adult steady state. Such a system exhibits a tendency towards a certain kind of equilibrium, which is restored after disturbance; but this equilibrium is not centred on a static state but rather on a direction or pathway of change. We might speak of such an equilibrium-property as a condition of 'homeorhesis', (ρηω, to flow) on the analogy with the well-known expression homeostasis, which is appropriate when it is an unchanging state which is maintained."
Super excited for your presentation, Professor! Do you think that there's some sort of "no-free lunch" concept in feedback control/homeostasis that makes it impossible to protect against all adversarial perturbations?
Yes, the point being computational irreducibility: in general, being able to protect against all adversarial perturbations would require that the control module has a perfect map of the controlled system's internal states, including itself. (Chris Fields and Wolfram have formalized this in somewhat different ways.)
Viewing generalization in machine learning through the lens of homeostasis seems very intuitive actually. Especially because, if I understand correctly, integral control is a feedback mechanism where the system continuously adjusts its output based on the cumulative error over time, ensuring tight regulation around a set point, unlike proportional control, which reacts only to the present deviation.
The case has been made for viewing biological systems as technological to help bridge how we understand and implement nanotech in biological/organic systems.
I wonder if the reverse can be explored.
I also wonder whether we can use the lens of homeostasis and autonomic integral control to scale and understand how larger and more complex systems operate. Especially because a lot of the data running through these feedback mechanisms reflects human behavior
I just wish Cannon had called it homeodynamics, which appears like a better analogy for what’s happening. Better yet, William Butts (not a Bart Simpson reference!) use of the term homeorhesis better describes a system that returns to the same trajectory in phase space.
Whoa, this is very interesting! Is homeorhesis generally the result of some type of feedback in the system? Apologies, I'm not really sure how to formalize this question.
The term "homeorhesis" was coined by C.H. Waddington, it refers to control mechanisms used by biological organisms to remain close to a viable trajectory of development and growth throughout their lifetimes.
From Waddington's _The Strategy of the Genes_ (1957):
"We spoke earlier of trajectories in the phase space as converging on one or another definite end-point. As a matter of fact, if a process of embryonic development is disturbed, it usually returns to normality some time before reaching the adult condition. Its trajectory, that is to say, converges not merely to the normal end state, but to some earlier point on the path leading towards the steady state. This is well symbolised by the epigenetic landscape. If a ball, running down one of the valleys, were pushed partway up the hillside, it might well reach the valley bottom again before the slope of the valley flattens out as it reaches the adult steady state. Such a system exhibits a tendency towards a certain kind of equilibrium, which is restored after disturbance; but this equilibrium is not centred on a static state but rather on a direction or pathway of change. We might speak of such an equilibrium-property as a condition of 'homeorhesis', (ρηω, to flow) on the analogy with the well-known expression homeostasis, which is appropriate when it is an unchanging state which is maintained."
Thank you Maxim, you explained it better than I could 🙏
Super excited for your presentation, Professor! Do you think that there's some sort of "no-free lunch" concept in feedback control/homeostasis that makes it impossible to protect against all adversarial perturbations?
Yes, the point being computational irreducibility: in general, being able to protect against all adversarial perturbations would require that the control module has a perfect map of the controlled system's internal states, including itself. (Chris Fields and Wolfram have formalized this in somewhat different ways.)
Viewing generalization in machine learning through the lens of homeostasis seems very intuitive actually. Especially because, if I understand correctly, integral control is a feedback mechanism where the system continuously adjusts its output based on the cumulative error over time, ensuring tight regulation around a set point, unlike proportional control, which reacts only to the present deviation.
The case has been made for viewing biological systems as technological to help bridge how we understand and implement nanotech in biological/organic systems.
I wonder if the reverse can be explored.
I also wonder whether we can use the lens of homeostasis and autonomic integral control to scale and understand how larger and more complex systems operate. Especially because a lot of the data running through these feedback mechanisms reflects human behavior
You might be interested in this: https://arxiv.org/abs/2312.17723