Years in the making: 304 papers cited in a huge effort to characterise this emerging field that we hold dearly.
Read this paper to find out what we are doing, why and how, and why we think that ANNs are a central tool for understand brain computations.
#neuroconnectionism
Just finished my course “Machine Learning for Cognitive Computational Neuroscience”. Across 12 lectures (90 minutes each) and 10 workgroup sessions, we covered >100 papers (46% published in the past two years). The students (and I) learned a lot.
Here is what we covered: 1/
🥳 My lab was awarded 1.5M Eur by
@ERC_Research
for our
#ERCStG
proposal
#TIME
. Over the next five years we will take a deep dive into visual information sampling, extraction, and integration by combining neuroimaging, psychophysics and large scale machine learning. So excited!
🚨Major professional news: My lab will leave the Donders and transition to the Institute of Cognitive Science at
@UniOsnabrueck
, where I will take on the role of Full Professor for neuro-inspired Machine Learning. 1/4
New preprint alert! We show that predictive coding is an emergent property of input-driven RNNs trained to be energy efficient. No hierarchical hard-wiring required. A thread: 1/
Tweeps, I am hiring (big time) - please spread the word.
I am looking for three postdocs in the space of neuro-inspired ML and cognitive computational neuroscience to join us at the Institute of Cognitive Science (
@UniOsnabrueck
).
Here is why this is a great opportunity: 🧵
"The majority of good ideas [towards human-level AI] will come from academia.
Even if the most impressive applications come from industry."
- Yann LeCun
@ylecun
@willjharrison
@OpenAI
Not this one, no, but after a great suggestion by
@__init_self
, we tried to push it to its limits with something more creative. I am blown away to say the least.
Are you using DNNs in your work? Then our new paper may be of interest to you: "Individual differences among deep neural network models", now out in
@NatureComms
:
A quick run-through:
Thrilled to announce that I just accepted an offer from the
@DondersInst
@AI_Radboud
! I will join their faculty as Assistant Professor this fall.
Get in touch if you would like to work with me in the area of (visual) computational neuroscience and machine learning!
PSA: Academia can be a harsh place, but it does not have to be.
- Celebrate beautiful work and findings by others.
- Read a paper that you liked? Let the authors know.
- Remember: science is not a zero-sum game.
#ScienceTogether
#ForwardTogether
#AcademicChatter
What is your favourite "ANN contributing to brain science" paper? Bonus points for NOT VISION: language, motor control, audition, planning, RL,... anything goes.
Background: I am putting together a lecture series and noticed how incredibly biased my examples are.
Just stumbled upon an old letter to the editor in which we respond to the reviewer's request to cite 6 (!) of his/her own papers.
Adding 2/6 in round 1 was not enough. We were told that the reviewer cannot accept until all 6 are added.
DON'T DO THIS TO PEOPLE.
Now out in PNAS!
We combine MEG source reconstruction, RSA, and end-to-end deep neural network training to show that "Recurrence is required to capture the representational dynamics of the human visual system".
#deeplearning
#neuroscience
#neuroimaging
My whole timeline has become more or less useless and boring: it's all talk about chatGPT and some hastily coded-up startups trying to cash in - all sensible research discussion has gone to the background.
My son figured out that he can make the biggest mess, declare it "an experiment", and totally get away with it because HOW CAN I NOT ALLOW EXPERIMENTS IN THIS HOUSE?
Excited that our paper "Predictive coding is a consequence of energy efficiency in recurrent neural networks" is now out in
@Patterns_CP
! Much has changed since preprint, so check it out:
Work with the fantastic
@hellothere_ali
,
@nasiryahm
, and
@marcelge
!
The Cognitive Computing AI MSc program at the Donders is dropping reference letters from the application process. We hope that this will reduce bias and lead to fairer evaluations of people of diverse backgrounds.
The lab has been scooped repeatedly in the last few weeks. Positive: we operate in a space of ideas that people find interesting. Negative: boy it sucks.
Help me out hive mind: is there a good twitter list of people in the interdisciplinary space of ML/Comp Neuro/AI?
If not, feel free to suggest people in this thread, trying to diversify my timeline. Bonus poins for researchers from underrepresented groups.
Der Bundesbericht Wissenschaftlicher Nachwuchs 2021 ist da:
"Die durchschnittliche
#Promotionsdauer
beträgt 5,7 Jahre (ohne Humanmedizin/Gesundheitswissenschaften)." (S. 137f., )
#BuWiN2021
🚨HIRING🚨 We are looking for a postdoc in the space of machine learning and visual computational neuroscience to join my group at the institute of cognitive science (position for 3 years, can be extended). Official call here (more info in the🧵 below):
Tweeps, I am hiring (big time) - please spread the word.
I am looking for three postdocs in the space of neuro-inspired ML and cognitive computational neuroscience to join us at the Institute of Cognitive Science (
@UniOsnabrueck
).
Here is why this is a great opportunity: 🧵
Thrilled to see this out. We show that recurrent DNNs can outperform parameter-matched feedforward DNNs in large-scale object classification, trade off speed for accuracy, and better predict human reaction time data.
New from Spoerer and colleagues: "Recurrent neural networks can explain flexible trading of speed and accuracy in biological vision" -
@KriegeskorteLab
@TimKietzmann
Please advise: Which female PIs or senior researchers in AI or neuro-AI are located in the EU? Looking for inspiration for a keynote speaker outside my regular bubble.
Please RT for visibility. Thx!
I am speechless. The
@Radboud_Uni
executive board has sent me these amazing flowers to our home in Germany to congratulate me on the ERC. Thank you so much! This is leadership par excellence.
I was asked to review A WHOLE BOOK for a for-profit publisher. What do they offer in return?
"For well-prepared review reports submitted in a timely manner, we provide discount vouchers for future publications by MDPI."
Wow.
Okay, I'll bite. Letters of reference (for MSc students and likely beyond) are utter nonsense, often written by the students themselves and signed off by a professor who barely knows them. Worst of all: their very existence discriminates against minorities and amplifies biases.
🚨ERC Job Alert 🚨
Are you looking for a PhD position? Are you into AI and/or computational/cognitive neuroscience? Then consider becoming an ERC funded graduate student with us:
I am biased, but I think this is a great opportunity. 🧵
MSc programs at Radboud often require students to upload a photo of themselves as part of their application.
I have no idea what the looks of a student have to do with their academic potential. The Cognitive Computing MSc AI program will therefore not follow suit.
Due to time constraints, I would have had to fly in, give the talk, and fly out again - a bad ratio of CO2 vs time spent interacting with other scientists.
So a big thank you to the
@mrccbu
for experimenting with me and having me over digitally. Worked like a charm.
Now onto the final session, we’re discussing how computational models are useful for studying the ventral stream.
@TimKietzmann
is our first speaker and giving his talk via Skype!
Tried to argue with admin that a good coffee machine will increase the productivity of the team way more than a single additional GPU ever could.... and was 100% unsuccessful.
As some of you have asked, here is a link to my CCN keynote from this year.
Main topic: how we use ANNs as a language to describe, implement, and test theories of brain computations.
#neuroconnectionism
I'm curious. Who ever said vision is "solved"? We teach AI students in their very first semester that DNNs remain far from human performance/robustness.
This one figure is pretty much a final nail to the coffin of the preposterous idea that deep learning somehow "solved" vision. Paper:
Thx
@MaxALittle
for posting ()
We ran a 'meme contest' in my course "ML for Cognitive Computational Neuroscience".
Student submissions were... niche. They were nerdy, they were well informed, they were brilliant, and so very funny.
Have two examples, it's Friday.
#CoxiOS
Alright, could we all agree to show figures in the main text of preprints (even if you decide to play along with anachronistic policies for your journal submission)?
So I bought a whiteboard for my home-office and it's the BEST purchase in a long while.
Here is why: first, I cannot meet without drawing, and I now point the webcam at the board during online meetings to aid discussion. 1/3
After being scattered across Europe for the past 1.5 years, the lab is finally embarking on a joint retreat. I cannot overstate how much I am looking forward to this. Beautiful spot, too.
Told my wife that my current level of crazy busy "is just a phase". She could not stop laughing and responded that I've been saying this for 15 years.
Time to face reality I guess.
I present to you: our 4 year old.
Her: "I'm sorry that you have to work again daddy".
... door closes ...
... 10 seconds...
... head pokes back in ...
Whispering: "Don't worry, you don't have to work when you are dead".
You could not make this stuff up.
Bored of all the dogs in ImageNet (ILSVRC)? Meet ecoset: 1.5m images from 565 basic level categories (selected to be of importance to humans). A new benchmark for
#ComputerVision
,
#ML
, and
#DeepLearning
.
Dataset and pre-trained CNNs here:
This is brilliant: successful predictions of conference paper acceptance from paper appearance alone.
Hidden gem: algorithm predicts its own paper to be rejected without peer review.
Would anyone mind sharing their setups for having a (power)nap in the office? Asking for a friend with small children who have absolutely no respect for their parents' sleep.