Simone Rossi

I have an open PhD position to work on the development of advanced machine learning tools for the analysis of the cerebrovascular tree at EURECOM. Deadline to apply: Oct 1st. More info here: eurecom.fr/sites/default/…

@francoisfleuret @n_keriven @NeurIPSConf Not true. The problem was the change of plans from the PC with little notice. When we accepted the invitation, the discussion phase was almost the entire month of August (like last year). Only later, this was split into Reviewer2Author (1 week) and Reviewer2Reviewer (2 weeks).

@n_keriven @NeurIPSConf Why boasting about this? You break a contract you were not forced to accept, and hurt authors for whom it may be a big deal.

@jmaronasm The idea to limit the discussion to 1 week was stupid, especially if everyone is expected to always have time to interact with everyone else.

In other words. Interactive review period in summer is bullshit.

My conclusion with #NeurIPS2022 review period is that I prefer @aistats_conf . You answer all reviewers in just one page and don't waste/allocate time for unusefull/inexistent interaction. If your feedback is good reviewers discuss without the need of your interaction. Own exp.

@ruqi_zhang Hi Ruqi, I’d love to have a chat with you on SG-MCMC methods and Bayesian deep learning one of these days.

I'm at #ICML2022! Look forward to meeting old and new friends. Ping me if you want to meet up :)

After writing 60640 words for thesis and 77426 lines of Python with more than 9 years of GPU compute time for all the experiments, I guess it's now official!

Words fall short today

Great way to end the day! I’m honored do be selected as Outstanding Reviewer for #NeurIPS2021. I must say, reviewing 7 papers in a couple of weeks and discussing with fellow reviewers/authors, all during the summer break was not an easy task.

I should eventually make some on the thesis, but not today

@michiard @MaurizioFilip19 @edwinvbonilla Dimitrios Milios

But in all seriousness, well done to Ba-Hien for this fantastic work on priors and Bayesian autoencoders arxiv.org/abs/2106.06245

Nothing better than a glass of fine scotch to celebrate @NeurIPSConf decisions. Works for both cases

@BlackHC This time I feel also the matching paper/reviewer was worse. As a reviewer, I got assigned a couple of papers that I didn't bid for and not 100% aligned with my expertise. As an author, I have a feeling that a couple of reviewers were a bit too much unfamiliar with the topic.

I'm disappointed as a reviewer for NeurIPS when I see how other reviewers only spent 2-3 hours on a paper, self-declared, and wrote awful reviews. The load was high, but that can be flagged, and it's not fair towards other reviewers (and the authors) who put in the effort

@openreviewnet looks to be down. I wonder why could that be!

With a bit of delay, I'm thrilled and excited to announce that our tutorial on "Bayesian Inference for Deep Learning" has been accepted at @IJCAIconf 2021. Join us to discover some recent advancements in BDL! ijcai-21.org/tutorials/ @MaurizioFilip19

@alfcnz @KyleCranmer @Apple My macro for conditional is \;|\;, it gives a nicer spacing between variables and nothing gets overly squeezed

@KyleCranmer @Apple Feedback: the conditional bar is obtained with the command \mid. Also, \,\text{d} would be nicer.

I love that @Apple Keynote now has native LaTeX support, but why does the integral symbol need to be so ugly?

@mblondel_ml I can add one more: sane compilation time. Why should I wait 1/2 minutes per compilation while on a laptop the same document is done in ~15/20 seconds? And you also have no idea of what is going on there, just a spinning wheel!

My reasons for preferring git over overleaf for editing LaTeX: 1) use my favorite text editor 2) use my favorite PDF viewer (instantaneous refresh, beautiful font rendering) 3) real commits with a commit message (commit when you're ready, rollback commits, know who edited what)

Rant of the day to ML conferences, while preparing poster/video for AISTATS: can we please stop to normalize the 3-min talks? No useful message can be scientifically divulged in less than 3 minutes. It just becomes a sales pitch. #AISTATS #ICML #NeurIPS #ICLR

Touchdown confirmed. The #CountdownToMars is complete, but the mission is just beginning.

Maybe we have better tools for describing functions (or better, stochastic processes). Gaussian processes seems a great fit for the job! For more details, check out our work at the Symposium (bit.ly/Tran2021) and the 5-min talk (bit.ly/Tran2021Video). 2/2

It looks like "Priors for Bayesian NN" is kind of an hot topic at this year @aabi_org! And rightfully so. Choosing a prior on millions of parameters is definitely not the most user-friendly task for anyone trying to use BDL in practice. 1/2