I was of three minds,
Like a paper
On which there are three reviews.

#NeurIPS #EMNLP #AAAI

Learn about the NeurIPS 2023 Machine Unlearning Challenge https://www.fosslife.org/new-machine-unlearning-challenge #MachineLearning #AI #research #privacy #ethics #NeurIPS

First #NeurIPS submission. Wish me luck! #NeurIPS2023

MLOps for Compositional AI paper presentation at #NeurIPS Workshop on Challenges in Deploying and Monitoring Machine Learning Systems (#DMML), 2022

https://slideslive.com/38994256

#mlops #MachineLearning #slideslive #dataops #compositionalai #DataScience #enterpriseai

Happy to share that my paper on 'MLOps for Compositional AI' has been accepted at #NeurIPS 2022 Workshop on Challenges in Deploying and Monitoring Machine Learning Systems (DMML).
Paper: https://www.researchgate.net/publication/364603411_MLOps_for_Compositional_AI

A blog version of the paper is also available on @towardsdatascience
https://towardsdatascience.com/bridging-dataops-mlops-301f010caf30

#neurips2022 #mlops #compositionalai #artificialintelligence #machinelearning #dataops #dmml #paper #wiproai #enterpriseai

Great news: We updated the schedule of the #AUTOML23 conference s.t. the deadlines are better aligned with other big ML conferences such as #NeurIPS. The new plan is that you will get #AUTOML23 notifications definitely before the NeurIPS deadline (to make resubmissions possible). 🎉 More details at https://www.automl.cc

📣 What do we want?!
Discoverable, Accessible,
Homogeneous, Scalable,
Distributable and Maintainable #OpenData !!!!!
📣 When do we want it??
NOW!!

In case you missed our #NeurIPS Workshop poster on Large Transnational Datasets, you can have a peak now 🤓

📄 Full Paper: https://arxiv.org/abs/2210.07178

After #NeurIPs, a student was saying that even preprints are not fast enough for advances in AI. He had found code and an explanation for a tool relevant to his work on Twitter before the preprint was even posted!!

Why perform cross validation (CV) in #MachineLearning? To estimate the generalization error of a trained predictor. This paper uses the idea of a #ProperLoss (called Q-class). Then it covers CV, bootstrap, and Mallow's covariance penalties. It also covers #ConformalPrediction, which is newly popular because of Emanuel Candes' keynote at #NeurIPS 2022
https://doi.org/10.3390/stats4040063
The paper is also a good advertisement for Efron and Hastie's recent book.

Do #NeurIPS proceedings get pulled by Google Scholar? Not seeing our NeurIPS paper on my paper list.

Do #NeurIPS proceedings get pulled by Google Scholar? Not seeing our NeurIPS paper on my paper list.

Combing #languagemodels with #reinforcementlearning , it's the way forward to tame these breasts. Makes sense that it won best paper at #neurips

https://openreview.net/forum?id=YoS-abmWjJc

#dramatron from #neurips https://deepmind.github.io/dramatron/details.html

I'm still experimenting with my @arxiv bot, as it looks like there aren't that many papers being posted on a regular basis now #NeurIPS is over... Since I don't want to retweet just the same two account over and over, the bot will likely go directly to the source feed to find cool papers using keyword filters.

Do we need one definition of Explainable AI to rule them all?

No, not right now. Why? This is where bicycles come in.

What am I talking about?
Join me & @Riedl at the Human-centered AI (HCAI) workshop at @NeuripsConf to learn more!

We argue why a singular definition of XAI is neither feasible nor desirable at this stage of XAI's development.

But why? A thread

📜https://arxiv.org/abs/2211.06499

#AI #ExplainableAI #XAI #NeurIPS #HCI #algorithms #bicycles

1/6

I presented our winning solution to the molecular property prediction competition at Neurips last night. Dataset was clean and very interesting. I hope they will run this competition again next year! Our technical report https://arxiv.org/abs/2211.11035 #machinelearning #neurips

I'll be at the #NeurIPS panel session at 6pm PT today with @ZejiangS@twitter.com discussing Pile of Law and MultiLexSum!

Come check it out at Panel 4C-5: https://neurips.cc/virtual/2022/session/65426

Pile of Law: https://arxiv.org/abs/2207.00220

Multi-LexSum: https://multilexsum.github.io

Extending datasets to ensure 50% women provably, offers the biggest bang (diversity) for buck (data points) in bringing diversity to our datasets. But more importantly to our models in ways that are less obvious but makes it literally the most valuable thing we can do if we want to be fairer, to everyone.
#ethicalai #responsibleai #fairness #Neurips

Of all the sensitive features one might consider in modelling fairness, sex /gender has the largest subgroups (assuming our population to be humans). As a result, if we just extend our datasets to include women, this will clearly have the largest impact. But the benefit would not be exclusively to women. When we build models, even if the data is binary, the model representation need not treat them as such, take for example logistic regression.
#ethicalai #responsibleai #fairness #Neurips

@chespace @Memotv well, it's both boring and racist. https://twitter.com/spiantado/status/1599462375887114240

I'm still unsure whether it could be useful at all. Considering it as a chat bot, my tests gave me ALICE-style useless generality.

People claim that it's good at copy-paste coding (Collab-style), yet as someone comfortable with software development I don't find this particularly useful. Then, there's all the Stack Overflow stuff happening. https://twitter.com/mapto/status/1600339199995125761

Tell me something worthy about it? Otherwise, I'll keep thinking that it, along with Galactica, is just #NeurIPS -timed marketing that has little relevance beyond a particular zealot-like crowd.

More generally, I've been recently playing with text-to-image. Much more satisfied by #Midjourney than DALL-E. The whole size-matters narrative seems more and more like a stunt to me.