David Lake

For a 100 m precision tag of a camera pose for lookup over a limited region, not much to concatenate the literal human interpretable hash. Tagging sets of features with ortho images or prior in situ images is ok, using global/aggregate features to get a pose prior in that sense is clear, but the bag of visual words model suggests we might not need that structure for a weaker estimate of position using the positions of query features and the features themselves in inverted file sense (with some conditions that features aren’t completely uninformative). Clustering on position metadata seems reasonable but curious if you have seen anything similar I could read

Using geohash or similar, more sophisticated spatial embeddings from libraries such as S2 to augment feature vectors is not new but this is a problem I like so a question for you. Is there a spatial stat / closed form solution to inferring (based on image query of features only ideally) position estimate based on the occurrences of features or groups of features and an inferred precision (radius)?

Google and ETH have joined the large scale localization effort with a banger I really did not expect this And now I'm really hoping to make it to NeurIPS where the paper will be presented I'll read it and report a summary here in the next few days

@Pantojamma humbled the new kid on the block, took his best shots then took him to deep water and made him look like a damn white belt. #UFC310 #Naptime

This is the most flowers fighters given out to each other in a night before. It was like a gd reverse roast @BCampbell . Dustin picks Conor over Chander. Jones #p4p4l according to Dana. Strickland picks Izzy over Dricus. Everyone agrees the judges suck #ufc302

@Cooper_burke9 Yeah, he got a professional courtesy type slow rollout from a position that required a power move to break the grip, a last warning extension, eyes to the ref to ensure he had his attention then finally Kevin snapped that shit off

Wow. Oleksiejczuk just broke his arm. Holland caught him in an armbar after the knockdown and he didn’t tap. What a wild moment, I feel like you have to know when to protect yourself. #UFC302

@TheFightGIF @espnmma Booped him there

@espnmma Crazy

Diego Ferreira snapped Mateusz Rebecki’s 16-fight win streak 💪 #UFCStLouis

@predict_addict Elements of statistical learning is too old. Not that the concepts didn’t exist but they weren’t considered must haves at the time. Intro to statistical learning is the friendly version & can understand why the authors might omit; also get why it would be good to include here

Scoop: the popular data science book '"The Elements of Statistical Learning" has exactly ZERO mentions of not only conformal prediction, but also "calibration" nor most important calibration metrics such as "Brier score". Let that sink in, one of the most popular books in data science has totally bypassed/ignored the concept of classifier calibration. #DataScience

@anmlbhvr7 @DovySimuMMA Exactly. Bobby took someone who otherwise wouldn’t have given him an extra 20 piece and tried to punk. Then jalin leaned in

@DovySimuMMA Good, maybe he’ll quit running his mouth off

bro Bobby green got fucked twitter.com/TSV__1/status/…

@HeyyUniquee Then got that work put on him

@HeyyUniquee He made the decision to try an punk jalin

@TheLastHokage11 Karma fired back at bobby bih quick

The Ref heard Bobby say What's Up and took it personally 😭 #UFCAustin

@GeneRogersVE Also jalin after bobby tried to punk him

@SpinninBackfist Hey love bobby but after that shit talking jalin was gonna go till the ref pulled him off

JALIN TURNER ON SHORT NOTICE KNOCKS OUT BONNY GREEN THREE TIMES #UFCAustin twitter.com/TSV__1/status/…

@_jasonwei Don't think he necessarily took it personally. It came across as a random declarative on my feed that is improved by the added context that sort of shows that the hypothesis actually just doesn't apply here. Nothing to see.

I tried to give this talk in the spirit of "a college soccer player watches videos of Messi and analyzes what makes him such a great soccer player." IMO it's great for people to aspire for greatness in AI research, just like in sports. Sorry you took it personally.

@cutezu_ That is interesting. I had always assumed it was called noise because it is essentially a combinatorial artifact that is hard to separate from the data in this case, not that it is non structured

@cutezu_ That there is structure is not unexpected given the combinatorics of cycles over different dims from the singular homology perspective, but I am now realizing that you might be discussing the nuances of how this changes with the filtration.

@cutezu_ The part of TDA I use is basically barcodes as a universal invariant/representation for data exploration, level set methods, and tools mostly toward Morse and more geometric bent, fwiw. Great post, will have to check up on more recent TDA and see if my hot takes still hold

@cutezu_ There are many graph/spectral approaches that currently much faster, provide similar insights, and have um.. fewer crack pot methods parading as valid theoretically sound tools. Again not saying it shouldn't be a goal to assess/develop the theory and methods :)