Mitchell Ostrow
@neurostrow.bsky.social
PhD Student at MIT Brain and Cognitive Sciences studying Computational Neuroscience / ML. Prev Yale Neuro/Stats, Meta Neuromotor Interfaces
Thanks again to all my amazing collaborators, especially my co-first author @annhuang42.bsky.social !
November 10, 2025 at 4:16 PM
Thanks again to all my amazing collaborators, especially my co-first author @annhuang42.bsky.social !
Public code is here github.com/mitchellostr... , and it is soon to be merged into the DSA package (pip install dsa-metric)
GitHub - mitchellostrow/DSA at inputdsa
Dynamical Similarity Analysis code accompanying the paper "Beyond Geometry: comparing the temporal structure of computation in neural circuits via dynamical similarity analysis" - GitHub ...
github.com
November 10, 2025 at 4:16 PM
Public code is here github.com/mitchellostr... , and it is soon to be merged into the DSA package (pip install dsa-metric)
Second, we develop a new similarity metric based in control theory and shape metrics, which is extremely fast and robust (no figure here)! The metric is based on controllability, which measures how easily inputs can arbitrarily move the state of a dynamical system.
November 10, 2025 at 4:16 PM
Second, we develop a new similarity metric based in control theory and shape metrics, which is extremely fast and robust (no figure here)! The metric is based on controllability, which measures how easily inputs can arbitrarily move the state of a dynamical system.
First, we apply subspace id methods from classical control theory to learn input-controlled linear dynamical systems (key in partially observed settings). This is new for the Dynamic Mode Decomposition (DMD) literature, and the method robust to extreme partial observation (12/)
November 10, 2025 at 4:16 PM
First, we apply subspace id methods from classical control theory to learn input-controlled linear dynamical systems (key in partially observed settings). This is new for the Dynamic Mode Decomposition (DMD) literature, and the method robust to extreme partial observation (12/)
Now for the 🤓 : InputDSA leverages 2 new technical developments (11/)
November 10, 2025 at 4:16 PM
Now for the 🤓 : InputDSA leverages 2 new technical developments (11/)
We think that inputDSA could be especially useful when experimentalists can perturb a system (e.g with optogenetics) for system identification. (10/)
November 10, 2025 at 4:16 PM
We think that inputDSA could be especially useful when experimentalists can perturb a system (e.g with optogenetics) for system identification. (10/)
As with DSA, inputDSA complements other comparison metrics (@itsneuronal.bsky.social , @mschrimpf.bsky.social ). One important result we found is that even for input-driven dynamics, the original DSA still gives good comparisons, but inputDSA can sharpen them! (9/)
November 10, 2025 at 4:16 PM
As with DSA, inputDSA complements other comparison metrics (@itsneuronal.bsky.social , @mschrimpf.bsky.social ). One important result we found is that even for input-driven dynamics, the original DSA still gives good comparisons, but inputDSA can sharpen them! (9/)
On two datasets, we apply random perturbations (noise, functions) to the true input, or utilize other task variables when performing inputDSA. We measure the correlation between the surrogate and true scores, finding that in general, inputDSA is quite robust! (8) (shoutout @oliviercodol.bsky.social)
November 10, 2025 at 4:16 PM
On two datasets, we apply random perturbations (noise, functions) to the true input, or utilize other task variables when performing inputDSA. We measure the correlation between the surrogate and true scores, finding that in general, inputDSA is quite robust! (8) (shoutout @oliviercodol.bsky.social)
One more analysis with greater implications: In most neuroscience settings, we don’t know the true inputs to a brain region. When we build models, we apply proxy inputs that we think are related to the true input. With InputDSA, we can evaluate this! (7/) (as in e.g line attractors in hypothalamus))
November 10, 2025 at 4:16 PM
One more analysis with greater implications: In most neuroscience settings, we don’t know the true inputs to a brain region. When we build models, we apply proxy inputs that we think are related to the true input. With InputDSA, we can evaluate this! (7/) (as in e.g line attractors in hypothalamus))
Second: on @thomas-zhihao-luo.bsky.social recently showed that rat cortical dynamics transition from primarily input-driven to autonomous during a 2-alternative forced choice task. InputDSA corroborates this, showing that cortex becomes less input-controllable across time! (6/)
November 10, 2025 at 4:16 PM
Second: on @thomas-zhihao-luo.bsky.social recently showed that rat cortical dynamics transition from primarily input-driven to autonomous during a 2-alternative forced choice task. InputDSA corroborates this, showing that cortex becomes less input-controllable across time! (6/)
On @satpreetsingh.bsky.social ’s Deep RL fly navigation task (from @bingbrunton.bsky.social ’s lab) we show that successful models become more similar to each other across training, while unsuccessful ones diverge in inputDSA score —an Anna Karenina/universality result! (5/)
November 10, 2025 at 4:16 PM
On @satpreetsingh.bsky.social ’s Deep RL fly navigation task (from @bingbrunton.bsky.social ’s lab) we show that successful models become more similar to each other across training, while unsuccessful ones diverge in inputDSA score —an Anna Karenina/universality result! (5/)
Let’s look at some cool applications first! We made a lot of technical developments, but I'll save those till the end 🤓 :
November 10, 2025 at 4:16 PM
Let’s look at some cool applications first! We made a lot of technical developments, but I'll save those till the end 🤓 :
The basic idea of DSA: approximate your dynamics so that comparison is tractable. This is backed by Koopman Operator Theory and relates to work done by @wtredman.bsky.social and Igor Mezic. InputDSA naturally extends DSA—we can compare intrinsic dynamics, the effect of input, or both jointly! (3/)
November 10, 2025 at 4:16 PM
The basic idea of DSA: approximate your dynamics so that comparison is tractable. This is backed by Koopman Operator Theory and relates to work done by @wtredman.bsky.social and Igor Mezic. InputDSA naturally extends DSA—we can compare intrinsic dynamics, the effect of input, or both jointly! (3/)
We introduce InputDSA, a method that builds on our prior work, Dynamical Similarity Analysis (DSA) to quantitatively compare input-drive dynamical systems! Especially relevant for neuroscience, but it can be applied to any type of time series data ! 🧠 💻 🌴 💨 💵 🔥 (2/)
November 10, 2025 at 4:16 PM
We introduce InputDSA, a method that builds on our prior work, Dynamical Similarity Analysis (DSA) to quantitatively compare input-drive dynamical systems! Especially relevant for neuroscience, but it can be applied to any type of time series data ! 🧠 💻 🌴 💨 💵 🔥 (2/)
This doesn't say anything about how the attractors is instantiated, ie the equation itself (let alone its mapping to the biology, which is another criterion needed for a mechanism according to Craver). I'm fine with this claim if it's what the post means!
July 9, 2025 at 2:59 AM
This doesn't say anything about how the attractors is instantiated, ie the equation itself (let alone its mapping to the biology, which is another criterion needed for a mechanism according to Craver). I'm fine with this claim if it's what the post means!
Perhaps what is meant by 'attractors aren't mechanisms' is that you can write down a large number of equations that are attractors (e.g. any diffeomorphism phi that transforms the system dxdt = -x while preserving its asymptotic behavior, also known as a conjugacy).
July 9, 2025 at 2:59 AM
Perhaps what is meant by 'attractors aren't mechanisms' is that you can write down a large number of equations that are attractors (e.g. any diffeomorphism phi that transforms the system dxdt = -x while preserving its asymptotic behavior, also known as a conjugacy).
I should clarify—in my example the system is the linear dynamical equation. When you say system, what are you referring to?
July 8, 2025 at 9:04 PM
I should clarify—in my example the system is the linear dynamical equation. When you say system, what are you referring to?
I guess I’m still not following, especially wrt usage of the words caused, composition and organization. Maybe we can use an example? Take dxdt=-x. the dynamics in this system is the decay to the attractor. So I’m not clear how to distinguish the two…
July 8, 2025 at 8:57 PM
I guess I’m still not following, especially wrt usage of the words caused, composition and organization. Maybe we can use an example? Take dxdt=-x. the dynamics in this system is the decay to the attractor. So I’m not clear how to distinguish the two…
I don't completely follow the second claim--it seems to me that there's no clear hierarchy between a system's behavior and the system, they are one and the same. So why aren't your two statements equivalent? Or are you talking about data--trajectories, and downstream inferences from it--attractors?
July 8, 2025 at 7:59 PM
I don't completely follow the second claim--it seems to me that there's no clear hierarchy between a system's behavior and the system, they are one and the same. So why aren't your two statements equivalent? Or are you talking about data--trajectories, and downstream inferences from it--attractors?
among other things relevant to arguing for ring attractor-ness, just thought that was most relevant to your article.
July 8, 2025 at 3:32 PM
among other things relevant to arguing for ring attractor-ness, just thought that was most relevant to your article.
Curious why you didn't include Chaudhuri et al. (2019)? They show that flows from perturbations off the ring manifold are biased back onto the ring (your section on 'Missing Activities')
July 8, 2025 at 3:31 PM
Curious why you didn't include Chaudhuri et al. (2019)? They show that flows from perturbations off the ring manifold are biased back onto the ring (your section on 'Missing Activities')
So no hotdog bowls? (I tried to find the Detroiters clip but couldn't someone pls link)
July 3, 2025 at 1:57 PM
So no hotdog bowls? (I tried to find the Detroiters clip but couldn't someone pls link)