sardean
@sardean.bsky.social
asst prof @cornellbowers.bsky.social thinking about dynamics, control, machine learning
sdean.website
sdean.website
almost a decade! how can that be true when I was almost certainly in high school or middle school a decade a go. (I will not be performing subtraction to verify)
January 22, 2026 at 5:06 PM
almost a decade! how can that be true when I was almost certainly in high school or middle school a decade a go. (I will not be performing subtraction to verify)
for some reason, Stephen Wolfram's walking desk strategy has not caught on despite his amazing 2019 blog post writings.stephenwolfram.com/2019/02/seek...
January 14, 2026 at 2:35 PM
for some reason, Stephen Wolfram's walking desk strategy has not caught on despite his amazing 2019 blog post writings.stephenwolfram.com/2019/02/seek...
the idea of "deli papers" being an issue is also new to me! The type of "thinly sliced contribution" that I expected to read about is the "minimum publishable unit" approach to scoping papers and projects. But that's distinct from submitting multiple "slices" at once. (Also probably harder to solve)
January 9, 2026 at 2:31 PM
the idea of "deli papers" being an issue is also new to me! The type of "thinly sliced contribution" that I expected to read about is the "minimum publishable unit" approach to scoping papers and projects. But that's distinct from submitting multiple "slices" at once. (Also probably harder to solve)
the slip from "this thing causes problems for some people" to "everyone is better off avoiding this" is so prevalent in baby safety recommendations. I'm not sure it's always wrong! But it's definitely not always right.
January 6, 2026 at 4:35 PM
the slip from "this thing causes problems for some people" to "everyone is better off avoiding this" is so prevalent in baby safety recommendations. I'm not sure it's always wrong! But it's definitely not always right.
I'm curious why SVMs in particular as the place for introducing kernels to ML students - what about least squares/ridge regression?
January 5, 2026 at 5:58 PM
I'm curious why SVMs in particular as the place for introducing kernels to ML students - what about least squares/ridge regression?
Fwiw, I don't think the article refutes a pro-Sacks view! I read it before seeing any discourse and didn't find it to be inflammatory or damning in any way. Just more insight into Oliver Sacks as a person.
December 13, 2025 at 3:19 PM
Fwiw, I don't think the article refutes a pro-Sacks view! I read it before seeing any discourse and didn't find it to be inflammatory or damning in any way. Just more insight into Oliver Sacks as a person.
github.com
October 24, 2025 at 2:55 PM
this includes Markov chains, Kalman filtering, optimal linear-quadratic control... can't seem to get away from linear models and quadratic costs
October 24, 2025 at 2:54 PM
this includes Markov chains, Kalman filtering, optimal linear-quadratic control... can't seem to get away from linear models and quadratic costs
Given the NSF funding priorities, everything's got to be AI these days. That or quantum I guess.
August 23, 2025 at 11:30 PM
Given the NSF funding priorities, everything's got to be AI these days. That or quantum I guess.
And in early August, balloon #2 ended it's journey in the Black Sea 🫡
August 18, 2025 at 2:52 PM
And in early August, balloon #2 ended it's journey in the Black Sea 🫡
Actually, I launched two balloons. But the first one crashed into the ocean off the coast of Nova Scotia due to a ballast issue. RIP
August 1, 2025 at 12:03 AM
Actually, I launched two balloons. But the first one crashed into the ocean off the coast of Nova Scotia due to a ballast issue. RIP
E.g. at netflix "we initially did not observe significant improvements in performance over well-tuned non-deep-learning approaches. Only when we added numerous features of heterogeneous types to the input data, deep-learning models did start to shine in our setting" ojs.aaai.org/aimagazine/i...
Deep Learning for Recommender Systems: A Netflix Case Study
| AI Magazine
ojs.aaai.org
July 21, 2025 at 2:31 PM
E.g. at netflix "we initially did not observe significant improvements in performance over well-tuned non-deep-learning approaches. Only when we added numerous features of heterogeneous types to the input data, deep-learning models did start to shine in our setting" ojs.aaai.org/aimagazine/i...
But more generally, this seems like a repeated pattern for deep learning. One of the main advantages is how flexible the models are, and how many disparate types of inputs they can be trained to use. This is in contrast to physics based or simpler ML models.
July 21, 2025 at 2:30 PM
But more generally, this seems like a repeated pattern for deep learning. One of the main advantages is how flexible the models are, and how many disparate types of inputs they can be trained to use. This is in contrast to physics based or simpler ML models.
however, I do think the deep models are promising, particularly because its easy for non-experts to incorporate new data sources as inputs. Credit to my brother for this perspective! www.nytimes.com/2025/07/13/b...
The Future of Weather Prediction Is Here. Maybe.
www.nytimes.com
July 21, 2025 at 2:29 PM
however, I do think the deep models are promising, particularly because its easy for non-experts to incorporate new data sources as inputs. Credit to my brother for this perspective! www.nytimes.com/2025/07/13/b...
they also rely on the initial conditions/"state estimates" provided by physics based models
July 21, 2025 at 2:26 PM
they also rely on the initial conditions/"state estimates" provided by physics based models
my (1000 foot view) of recent advances in deep weather models is that they basically get better accuracy metrics by doing a better job of being Bayes optimal. I.e., predicting the mean of future outcomes (vs more realistic looking high resolution weather patterns).
July 21, 2025 at 2:25 PM
my (1000 foot view) of recent advances in deep weather models is that they basically get better accuracy metrics by doing a better job of being Bayes optimal. I.e., predicting the mean of future outcomes (vs more realistic looking high resolution weather patterns).
basically, there are still wins from giving the models better initial conditions. (eventually that will saturate because of the chaotic dynamics/numerical precision/etc)
July 21, 2025 at 2:23 PM
basically, there are still wins from giving the models better initial conditions. (eventually that will saturate because of the chaotic dynamics/numerical precision/etc)
they suggest that there are still gains to be made by collecting/incorporating more real time weather data, specifically that we could get than currently usable predictions several more days into the future
July 21, 2025 at 2:22 PM
they suggest that there are still gains to be made by collecting/incorporating more real time weather data, specifically that we could get than currently usable predictions several more days into the future
I've been getting interested in weather prediction, where (depending on how much you believe physics models) there are some interesting things you can say: journals.ametsoc.org/view/journal...
journals.ametsoc.org
July 18, 2025 at 9:22 PM
I've been getting interested in weather prediction, where (depending on how much you believe physics models) there are some interesting things you can say: journals.ametsoc.org/view/journal...
totally agree, the market of baby products feels pretty scammy (not to mention all the targeted ads). But the same anxieties that all these products prey on seem to be reinforced by the safety culture...or at least this is the case on reddit forums full of anxious new parents.
March 1, 2025 at 10:52 PM
totally agree, the market of baby products feels pretty scammy (not to mention all the targeted ads). But the same anxieties that all these products prey on seem to be reinforced by the safety culture...or at least this is the case on reddit forums full of anxious new parents.