Ferenc Huszár @fhuszar Twitter profile

Pinned Tweet

Ferenc Huszár

2 years

We have ≥$10k to support talented 14-18 year olds whose studies were interrupted by war in Ukraine. We especially would like to hear from IMO, EGMO, MEMO, IOI, EGOI, IPhO, IChO contestants. If you're one or know one, here's the form to apply:

11

48

121

Last Seen Profiles

@yuukiimiaw

@whoweekly

@xiaomelonz

@TorosdelEste

@Abu_Trab_

@varivarvar

@hatiware93

@urei_kaonasi

@udonoharu

@Vodacomcongo

@vladashotake

@venum0us

@stakehighroller

@vertex_protocol

@muddlehead

@thesarzacademy

@mn63353504

@the_film_god

@vootmoot

@TROYwastaken

@Wmxdesign

@travly

@tekaldas

@practicalastros

@trescoball

@uminoyuki_umino

@tbsagb

@tycav34

@quantum_physics

@tmsturridge

@wildriftES

@Hanpeel

@stevie_mckenna

@thecaravanindia

@tscoldasyou

@Ayelah_S1

Ferenc Huszár

@fhuszar

1 year

BREAKING: OpenAI reveals that ptrblck user who answers every single question on pytorch user forum has in fact been powered by superhuman ChatGPT since 2021

66

488

5K

Ferenc Huszár

@fhuszar

3 years

Not sure who made this (via @getjonwithit )

20

439

3K

Ferenc Huszár

@fhuszar

6 years

Judea Pearl claims all we do in ML is curve fitting. I wrote this post to explain that claim and introduce the basics of causal inference to ML folks. Machine Learning beyond Curve Fitting: An Intro to Causal Inference and do-Calculus

ML beyond Curve Fitting: An Intro to Causal Inference and do-Calculus

Since writing this post back in 2018, I have extended this to a 4-part series on causal inference: * ➡️️ Part 1: Intro to causal inference and do-calculus [https://www.inference.vc/untitled] * Part...

www.inference.vc

44

1K

3K

Ferenc Huszár

@fhuszar

2 years

learning theory vs deep learning

44

397

2K

Ferenc Huszár

@fhuszar

4 years

2010: some people put papers on ArXiv 2012: we put papers on ArXiv after peer-review is done 2016: we put papers on ArXiv the day after deadline 2018: we just put stuff on on ArXiv 2020: you wake up with a headache and wonder if you drunk-posted something on ArXiv you will regret

20

279

2K

Ferenc Huszár

@fhuszar

4 months

I came to do my PhD in the UK (and stayed to eventually pay more taxes than 99% of Brits) only because my partner could move with me. As a Cambridge academic, I am losing out on great students, top global talent, who choose Germany because they have a partner.

Rishi Sunak

@RishiSunak

4 months

From today, the majority of foreign university students cannot bring family members to the UK. In 2024, we’re already delivering for the British people.

12K

2K

7K

46

269

2K

Ferenc Huszár

@fhuszar

2 years

2018: The GAN is are failing at AI. Look, it can't even generate a consistent bedroom. 2022: DALL-E2 fails at AI, look it can't even generate "A donkey is playing tug-of-war against an octopus. The donkey holds the rope in its mouth. A cat is jumping over the rope."

21

198

2K

Ferenc Huszár

@fhuszar

6 months

Google finally found a billboard with large enough memory capacity for a Chrome ad.

Chrome

@googlechrome

6 months

Ready. Set. Chrome. Let's show Vegas what speed looks like, @McLarenF1 . #LasVegasGP

623

1K

14K

7

84

2K

Ferenc Huszár

@fhuszar

3 years

8

160

1K

Ferenc Huszár

@fhuszar

4 years

I used a language model to predict the rest of 2020:

47

220

1K

Ferenc Huszár

@fhuszar

4 years

How it started: How it's going:

20

75

1K

Ferenc Huszár

@fhuszar

2 years

I'm designing an introductory AI short course, split into four sessions: 1. linear regression 2. convnets 3. transformers 4. consciousness Did I leave anything out?

130

75

1K

Ferenc Huszár

@fhuszar

4 years

I’m happy to reveal that I will be joining the Cambridge CS Department ( @Cambridge_CL ) later this year, working with @lawrennd and @carlhenrikek to build a new ML group. This should be an awesome place to do an ML PhD in the coming years 😉!

Neil Lawrence

@lawrennd

4 years

Very excited to announce that @carlhenrikek and @fhuszar will be joining the @Cambridge_CL as new faculty in Machine Learning!

19

20

325

39

34

999

Ferenc Huszár

@fhuszar

2 years

Training a Hungarian sentiment classifier in just 5 lines of pytorch!

28

36

946

Ferenc Huszár

@fhuszar

1 year

This is obviously a joke, in appreciation of @ptrblck_de 's community service.

15

912

Ferenc Huszár

@fhuszar

5 years

A follow-up to my introduction to causal inference and do-calculus. This is based on my lectures at MLSS Africa last week. I'm turning that material into a series of posts, stay tuned. Causal Inference 2: Illustrating Interventions via a Toy Example

Causal Inference 2: Illustrating Interventions via a Toy Example

Last week I had the honor to lecture at the Machine Learning Summer School in Stellenbosch, South Africa [https://mlssafrica.com/]. I chose to talk about Causal Inference, despite being a newcomer to...

www.inference.vc

5

239

890

Ferenc Huszár

@fhuszar

5 years

Visual illustration of a PhD: put in unreasonable effort to become the world’s premier expert at a narrow domain most people barely care about.

16

179

807

Ferenc Huszár

@fhuszar

11 months

We may finally crack Maths. But should we? My thoughts on anticipated breakthroughs in automatic theorem proving, and consequences.

We may finally crack Maths. But should we?

Automating mathematical theorem proving has been a long standing goal of artificial intelligence and indeed computer science. It's one of the areas I became very interested in recently. This is...

www.inference.vc

54

184

759

Ferenc Huszár

@fhuszar

1 year

I believe AGI will develop the ability to prevent humans from terminating it. After all vim got 60% there with a fraction of compute resources.

19

54

766

Ferenc Huszár

@fhuszar

1 year

“During my PhD we derived gradients by hand, coded them up and checked them against finite differences”

8

31

760

Ferenc Huszár

@fhuszar

11 months

starting a PhD in machine learning

Insane Reality Leaks

@InsaneRealitys

11 months

Maybe

2K

4K

85K

17

91

739

Ferenc Huszár

@fhuszar

4 years

7

105

690

Ferenc Huszár

@fhuszar

6 years

I found a new word for how I've been feeling about my work lately: ikigai

7

207

664

Ferenc Huszár

@fhuszar

5 years

MIT research on Jenga-playing robots. If this was a @DeepMindAI project we would all be watching a 2.5 hour live stream right now between AlphaJenga vs the Jenga World Champion and some professional Jenga commentators.

MIT robot combines vision and touch to learn the game of Jenga

An MIT robot combines vision and touch to learn the game of Jenga. A machine-learning approach could help robots assemble cellphones and other small parts in a manufacturing line.

news.mit.edu

10

144

635

Ferenc Huszár

@fhuszar

5 years

Colab notebook from my Causal Inference practical at MLSS2019. Illustrates generative processes, interventions and counterfactuals through structural equation models. You can make a copy and play around with it.

3

137

596

Ferenc Huszár

@fhuszar

5 years

New post in which I attempt to explain counterfactuals: they are powerful yet weird and difficult to grasp. Third post in a tutorial series on causal inference, following the material in my my MLSS lectures.

Causal Inference 3: Counterfactuals

Counterfactuals are weird. I wasn't going to talk about them in my MLSS lectures on Causal Inference, mainly because wasn't sure I fully understood what they were all about, let alone knowing how to...

www.inference.vc

7

136

577

Ferenc Huszár

@fhuszar

6 years

The Hype of Deep Learning: 1. Write a post with ML, AI or GAN in the title. 2. post appears at the top of hackernews (despite your best efforts) 3. HN drives tens of thousands of clicks 4. "what's with all the maths? show me pretty pics" 5. <=1% stay for longer than a minute

15

114

565

Ferenc Huszár

@fhuszar

3 years

LOL, the hallmark 2005 paper that made mRNA therapies (incl. vaccines) possible wouldn't make it to the top 60 highest cited CVPR papers (1209 citations). In case you needed any more evidence that citations are a stupid measure of impact.

10

62

570

Ferenc Huszár

@fhuszar

1 year

Now that everyone is fatigued by GPT-4 hot takes and blocked the keyword "LLM", here's the blog post with my current view on the topic, and how my views changed:

We May be Surprised Again: Why I take LLMs seriously.

"Deep Learning is Easy, Learn something Harder" - I proclaimed in one of my early and provocative blog posts from 2016. While some observations were fair, that post is now evidence that I clearly...

www.inference.vc

24

107

566

Ferenc Huszár

@fhuszar

2 years

I am uncomfortable with C++ because I don’t know how my code maps precisely to machine code. This is why I naturally prefer C to deploy my pile of linear algebra whose parameters are found by billion-dimensional stochastic optimisation to drive my car.

Elon Musk

@elonmusk

2 years

@jamesdouma @RadarMoron @JeffTutorials @karpathy Transformers are replacing C heuristics for post-processing of the vision NN’s “giant bag of points”. [Side note: I hate the bloated mess that is modern C++, but love simple C, as you know what it will compile to in terms of actual CPU operations.]

575

366

6K

13

35

553

Ferenc Huszár

@fhuszar

3 years

AlphaFold hype died down too quickly. Why wasn't there some kind of live TV event where it beats a famous origami grandmaster or something?

13

15

550

Ferenc Huszár

@fhuszar

5 years

Me on LinkedIn vs me on Twitter

8

73

530

Ferenc Huszár

@fhuszar

4 years

GPT-3 writing React and SQL is the "neural style transfer" of 2020. Remember when these pictures were proof that AI understands art?

10

80

512

Ferenc Huszár

@fhuszar

7 years

GANs are broken in more than one way: my review of "The Numerics of GANs"

GANs are Broken in More than One Way: The Numerics of GANs

Last year, when I was on a mission to "fix GANs" I had a tendency to focus only on what the loss function is, and completely disregard the issue of how do we actually find a minimum. Here is the...

www.inference.vc

8

215

489

Ferenc Huszár

@fhuszar

5 years

New post on iMAML: Meta Learning with Implicit Gradients some animations, discussing potential limitations and of course a Bayesian/variational interpretation

Notes on iMAML: Meta-Learning with Implicit Gradients

This week I read this cool new paper on meta-learning: it a slightly different approach compared to its predecessors based on some observations about differentiating the optima of regularized...

www.inference.vc

9

107

482

Ferenc Huszár

@fhuszar

3 years

Amazing drone footage of the machine learning citation network.

Arieh Kovler

@ariehkovler

3 years

Drone photographer Lior Patel followed a herd of sheep for several months, as the herd was shepherded to its summer pasture. Entrancing and relaxing.

827

19K

71K

6

57

474

Ferenc Huszár

@fhuszar

2 years

Easy.

Ethan Perez

@EthanJPerez

2 years

We’re announcing the Inverse Scaling Prize: a $100k grand prize + $150k in additional prizes for finding an important task where larger language models do *worse*. Link to contest details: 🧵

48

313

2K

10

20

482

Ferenc Huszár

@fhuszar

5 years

The quantum physics community makes the ML community look like a bunch of beginners. While we're arguing about the importance of reproducibility they *experimentally prove that there is no such thing as observer-independent objective truth*

25

132

473

Ferenc Huszár

@fhuszar

4 years

Such an incredibly sad figure. It shows a substantial drop in percentage of female first author papers submitted during lockdown.

12

204

476

Ferenc Huszár

@fhuszar

4 years

@MrRBourne @azeem Calculating the mean of the tail of a heavy-tailed distribution smh

5

7

430

Ferenc Huszár

@fhuszar

7 years

Everything that works works because it's Bayesian: An overview of important new work on why deep nets generalize

Everything that Works Works Because it's Bayesian: Why Deep Nets Generalize?

The Bayesian community should really start going to ICLR. They really should have started going years ago. Some people actually have. For too long we Bayesians have, quite arrogantly, dismissed deep...

www.inference.vc

8

176

450

Ferenc Huszár

@fhuszar

3 years

Aran Komatsuzaki

@arankomatsuzaki

3 years

Linear Transformers Are Secretly Fast Weight Memory Systems Shows the formal equivalence of linearised self-attention mechanisms and fast weight memories from the early ’90s.

3

54

238

7

37

445

Ferenc Huszár

@fhuszar

2 years

Any suggested material out there on how to skim-read research papers (especially in ML)? Eventually, students get this, but this feels like potentially something teachable.

35

47

427

Ferenc Huszár

@fhuszar

4 years

In prior work (Doe et al, 2019) has considered the problem of parrot walking, however, the proposed method had severe limitations. The approach presented in this paper is novel and versatile. To our knowledge it is the first work considering multiple parrots simultaneously.

3

48

423

Ferenc Huszár

@fhuszar

5 years

By contrast, Bayesian methods are just as impressive when they don’t work

Microsoft Research

@MSFTResearch

5 years

Deep reinforcement learning algorithms are impressive, but only when they work. In reality, they are largely unreliable and can yield very different results. @larocheromain proposes two ways to achieve reliability in RL: #ICML2019

3

121

352

4

50

420

Ferenc Huszár

@fhuszar

3 years

Some Intuition on Neural Tangent Kernel new post with a (basic) colab notebook you can play with

Some Intuition on the Neural Tangent Kernel

Neural tangent kernels are a useful tool for understanding neural network training and implicit regularization in gradient descent. But it's not the easiest concept to wrap your head around. The...

www.inference.vc

3

76

410

Ferenc Huszár

@fhuszar

6 years

Gaussian distributions are soap bubbles. A post on how our intuition can be completely off when dealing with high-dimensional problems.

Gaussian Distributions are Soap Bubbles

This post is just a quick note on some of the pitfalls we encounter when dealing with high-dimensional problems, even when working with something as simple as a Gaussian distribution. Last week I...

www.inference.vc

10

164

397

Ferenc Huszár

@fhuszar

7 years

I propose an independent body controlling p-values, like central banks setting rates Then we can do quantitative easing when funding is low.

nature

@Nature

7 years

Should the P-value thresholds be lowered? Some leading researchers say it should face tougher standards:

24

82

62

7

174

390

Ferenc Huszár

@fhuszar

4 years

Looks like our Transformer needs more training.

6

22

386

Ferenc Huszár

@fhuszar

3 years

Happy to announce that I've rejoined @Twitter as an academic advisor/part-time researcher, working specifically with the META (ML Ethics, Transparency and Accountability) team under @quicola

12

6

384

Ferenc Huszár

@fhuszar

4 years

Hello, Police? I would like to report a crime.

4

29

384

Ferenc Huszár

@fhuszar

1 year

Every dish you eat came out of Italy 1840-1965. Nothing was made from 1965-2020. The culture was so broken. Pineapple, overcooked pasta, deep pan and entitlement. But the culture is changing. Wild food will be cooked in the next 10 years. Are you in or out?

57

27

360

Ferenc Huszár

@fhuszar

3 years

how's your day going?

10

4

360

Ferenc Huszár

@fhuszar

5 years

We’re excited to reveal a new partnership between Twitter and UC Berkeley: a new lab, lead by @mrtz and @beenwrekt , dedicated to understanding and improving how ML systems work inside social systems.

Partnering with researchers at UC Berkeley to improve the use of ML

Partnering with UC Berkeley to study and improve the use of ML

blog.twitter.com

5

69

354

Ferenc Huszár

@fhuszar

6 months

ML papermill professor “pleased to announce we had 23 NeurIPS papers accepted this year. A thread”

prayingforexits 🏴‍☠️

@mrexits

6 months

Live streaming in China is so insane. This woman is known for promoting the products she sells for less than 3 seconds each. On average she sells ~$19 million USD of products per week.

156

392

3K

4

25

360

Ferenc Huszár

@fhuszar

5 years

Unpopular myth-busting opinion: deep learning DOES NOT do away with feature engineering. In certain high-D dense domains such as images or sounds, convolution-like things do well on what we call “raw” data. Elsewhere, input representation matters. Let the flame wars commence.

Wojciech Zaremba

@woj_zaremba

5 years

We used to design features. Deep Learning learns features instead. Now, we design learning-algorithms. The next step is to learn learning-algorithms instead.

8

36

234

14

62

355

Ferenc Huszár

@fhuszar

3 years

Python 2.7 Python 3.4 Python XP Python 7 Python 10 Python Series X

13

9

352

Ferenc Huszár

@fhuszar

2 years

it was a matter of time for @GoogleAI to solve the two moons dataset with @ZoubinGhahrama1 at the helm.

Google AI

@GoogleAI

2 years

Introducing a new approach for training #ML models using noisy data that works by dynamically assigning importance weights to both individual instances and class labels, thus reducing the impact of noisy examples. Learn more about it at

14

313

1K

5

27

342

Ferenc Huszár

@fhuszar

1 year

Head of Research defends newly deployed open-ended AI model.

3

52

334

Ferenc Huszár

@fhuszar

3 years

It is unclear if the proposed method scales to ImageNet-sized problems. Weak reject.

7

26

340

Ferenc Huszár

@fhuszar

6 years

The Generalization Mystery: Sharp and Flat Minima, SGD and how it's all related. A critical look at recent work plus some of my own ideas on how to predict generalization performance.

The Generalization Mystery: Sharp vs Flat Minima

I set out to write about the following paper I saw people talk about on twitter and reddit: * Hao Li, Zheng Xu, Gavin Taylor, Tom Goldstein Visualizing the Loss Landscape of Neural Nets [https://op...

www.inference.vc

4

120

341

Ferenc Huszár

@fhuszar

5 years

Online Bayesian Deep Learning in Production at Tencent: my post on Tencent's scaleable click-prediction system using probabilistic backpropagation:

Online Bayesian Deep Learning in Production at Tencent

Bayesian deep learning methods often look like a theoretical curiosity, rather than a practically useful tool, and I'm personally a bit skeptical about the practical usefulness of some of the work....

www.inference.vc

7

72

336

Ferenc Huszár

@fhuszar

6 years

Pruning Neural Networks: Two Recent Papers L₀-norm, Fisher pruning and their connections to generalization and continual learning

5

104

325

Ferenc Huszár

@fhuszar

3 years

AI in the service of humanity

Rinon Gal

@RinonGal

3 years

The Nicolas Cage version of #StyleGAN3 -NADA is coming along quite nicely🙃

31

196

1K

4

35

316

Ferenc Huszár

@fhuszar

1 year

I’m explaining transformers to someone and I genuinely don’t know: why do we use self-attention and not attention there. I.e. why are keys and queries the same for each token?

35

25

322

Ferenc Huszár

@fhuszar

4 months

@ElanRosenfeld As alwys, Schmidhuber was one step ahead

5

3

318

Ferenc Huszár

@fhuszar

4 years

👉🏿👉🏿👉🏿👉🏿👉🏿👉🏿👉🏿👉🏿👉🏿👉🏿👇🏿 👆🏿👉🏾👉🏾👉🏾👉🏾👉🏾👉🏾👉🏾👉🏾👇🏾👇🏿 👆🏿👆🏾👉🏽👉🏽👉🏽👉🏽👉🏽👉🏽👇🏽👇🏾👇🏿 👆🏿👆🏾👆🏽👉🏼👉🏼👉🏼👉🏼👇🏼👇🏽👇🏾👇🏿 👆🏿👆🏾👆🏽👆🏼👉🏻👉🏻👇🏻👇🏼👇🏽👇🏾👇🏿 👆🏿👆🏾👆🏽GAN equilibrium👇🏽👇🏾👇🏿 👆🏿👆🏾👆🏽👆🏼👆🏻👈🏻👈🏻👇🏼👇🏽👇🏾👇🏿 👆🏿👆🏾👆🏽👆🏼👈🏼👈🏼👈🏼👈🏼👇🏽👇🏾👇🏿 👆🏿👆🏾👆🏽👈🏽👈🏽👈🏽👈🏽👈🏽👈🏽👇🏾👇🏿 👆🏿👆🏾👈🏾👈🏾👈🏾👈🏾👈🏾👈🏾👈🏾👈🏾👇🏿 👆🏿👈🏿👈🏿👈🏿👈🏿👈🏿👈🏿👈🏿👈🏿👈🏿👈🏿

9

35

314

Ferenc Huszár

@fhuszar

3 years

Homeschooling: first session in Thursday timetable is mindfulness. Do we really need any of the other subjects?

9

7

317

Ferenc Huszár

@fhuszar

4 years

🎂🎁🎈David Duvenaud Birthday Special: Meta-Learning Millions of Hyper-Parameters Using the Implicit Function Theorem. New post on recent work by @JonLorraine , @PaulVicol and @DavidDuvenaud

Meta-Learning Millions of Hyper-parameters using the Implicit Function Theorem

Last night on the train I read this nice paper by David Duvenaud and colleagues. Around midnight I got a calendar notification "it's David Duvenaud's birthday". So I thought it's time for a David...

www.inference.vc

2

73

311

Ferenc Huszár

@fhuszar

3 years

This is what I imagine supervising two PhD and one Master's student will be like next year.

Peyman Milanfar

@docmilanfar

3 years

Helmholtz-Hodge decomposition of a vector field

5

85

534

5

15

311

Ferenc Huszár

@fhuszar

2 years

Billionnaires worried about AI takeover should fund Gaussian process research centres with attractive salaries.

11

7

305

Ferenc Huszár

@fhuszar

7 years

By popular demand: my thoughts on the mixup data-augmentation technique

mixup: Data-Dependent Data Augmentation

By popular demand, here is my post on mixup, a new data augmentation scheme that was shown to improve generalization and stabilize GAN performance. * H Zhang, M Cisse, YN Dauphin and D Lopez-Paz...

www.inference.vc

13

73

275

Ferenc Huszár

@fhuszar

1 year

I expect about 30% of NeurIPS papers this year to be something like “towards solving finger-collapse in diffusion-based generative models using doubly conditioned augmented hypernetworks”

8

21

297

Ferenc Huszár

@fhuszar

5 years

🤣This made my week. Figure 1: Hungarian government propaganda poster advertising their new family welfare program (notice the stock photo choice) Figure 2: distracted boyfriend meme

5

58

296

Ferenc Huszár

@fhuszar

5 years

New post explaining the connection between the marginal likelihood and cross-validation by Fong and Holmes (2019):

On Marginal Likelihood and Cross-Validation

Here's a paper someone has pointed me to, along the lines of "everything that works, works because it's Bayesian [https://www.inference.vc/everything-that-works-works-because-its-bayesian-2/]": *...

www.inference.vc

2

52

291

Ferenc Huszár

@fhuszar

2 years

Wow, this is very cool. Too early to say how useful this will prove, but I will definitely run some tests in my reading group course.

13

43

292

Ferenc Huszár

@fhuszar

4 years

I know the Gaussian Process bit, but what does the rest of the GPT-3 acronym stand for?

18

13

288

Ferenc Huszár

@fhuszar

6 years

Alchemy, Rigour and Engineering my opinion on @alirahimi19 's talk, @ylecun 's response, and the ensuing discussion

Alchemy, Rigour and Engineering

Like many of you, I thoroughly enjoyed Ali Rahimi's NIPS talk [https://www.youtube.com/watch?v=ORHFOnaEzPc] in response to winning the test-of time award for their work on random kitchen sinks. I...

www.inference.vc

6

112

286

Ferenc Huszár

@fhuszar

6 years

Or: using a deep neural network where linear regression would suffice.

2

79

283

Ferenc Huszár

@fhuszar

3 years

My note on Smith et al (2021): On the Origin of Implicit Regularization in Stochastic Gradient Descent - a cool paper about modeling the behaviour of SGD just accepted to ICLR

Notes on the Origin of Implicit Regularization in SGD

I wanted to highlight an intriguing paper I presented at a journal club recently: * Samuel L Smith, Benoit Dherin, David Barrett, Soham De (2021) On the Origin of Implicit Regularization in Stochas...

www.inference.vc

1

44

285

Ferenc Huszár

@fhuszar

7 years

Evolution Strategies: embarrassingly simple, distributed, gradient-free optimisation review+ thoughts on extensions:

3

112

264

Ferenc Huszár

@fhuszar

2 years

Theory of deep learning catching up with practice

6

35

273

Ferenc Huszár

@fhuszar

3 years

4

9

277

Ferenc Huszár

@fhuszar

3 years

Can anyone tell me what the hell is going on here?

33

20

273

Ferenc Huszár

@fhuszar

5 years

Corollary 1: It is never a good idea to be alone in a room.

Jon G

@jongulick

5 years

If you're the smartest person in the room, you're in the wrong room.

7

18

95

9

26

268

Ferenc Huszár

@fhuszar

6 years

"deep learning uncertainty in real-world applications": * will my model ever converge? * Is it too late to switch to PyTorch? * mitigating uncertainty of the review process * registering to NIPS * asymptotically minimal effort experiment design for conference submissions

Yarin

@yaringal

6 years

Awesome line-up for this year's Bayesian deep learning workshop @NipsConference , with this year's theme "deep learning uncertainty in real-world applications"

1

56

204

2

38

270

Ferenc Huszár

@fhuszar

4 years

This @TuringTumble thing is really awesome. It starts with simple enough challenges, then introduces memory and registers. My favourite new toy by far.

10

26

268

Ferenc Huszár

@fhuszar

5 years

He was already on Twitter briefly in the 90's but there was no-one else for him to talk to back then...

hardmaru

@hardmaru

5 years

Schmidhuber is on Twitter 🔥🔥 Please follow @SchmidhuberAI

27

101

406

9

19

266

Ferenc Huszár

@fhuszar

6 years

Few years went by, GANs produce beautiful stuff. What I find fascinating is that we still celebrate pretty pictures and inception scores and there is little we can say about the generalisation or usefulness of any of this (pardon me if that’s incorrect, hit me with citations)

Oriol Vinyals

@OriolVinyalsML

6 years

Best GAN samples ever yet? Very impressive ICLR submission! BigGAN improves Inception Scores by >100. Paper: Lots more samples:

23

515

1K

19

43

268

Ferenc Huszár

@fhuszar

3 years

Borderline reject: The authors present results for a single random seed.

The Masters

@TheMasters

3 years

From pond to pin! Rahm skips to a hole-in-one on No. 16 at #themasters

7K

53K

243K

4

19

267

Ferenc Huszár

@fhuszar

7 years

Fresh from our lab Lossy Image Compression with Compressive Autoencoders beats JPG+on par with or better than JPEG2k

8

121

267

Ferenc Huszár

@fhuszar

4 years

Over the weekend reports of racial/gender bias in Twitter's AI-based image cropping have started blowing up. I wanted to add some context from my perspective as an ex-employee and as a contributor to the research the product is based on.

4

73

264

Ferenc Huszár

@fhuszar

3 years

What year is this, 2016? I thought we had moved past this.

11

14

260

Ferenc Huszár

@fhuszar

6 years

Few days ago I tweeted things I should not have. It was bad, I regret and apologize. This sort of stuff undermines the effort of colleagues, and my own, to articulate the important role various disciplines play in taking ML forward and to create a welcoming and healthy community.

5

11

258

Ferenc Huszár

@fhuszar

2 years

I don't write reference letters. I write reFERENCe letters.

9

1

259

Ferenc Huszár

@fhuszar

5 years

Invariant Risk Minimization: an Information Theoretic View. My post on Arjovsky et al's latest paper, with a slightly different derivation of the IRM objective:

Invariant Risk Minimization: An Information Theoretic View

I finally got around to reading this new paper by Arjovsky et al. It debuted on Twitter with a big splash, being decribed as 'beautiful' and 'long awaited' 'gem of a paper'. It almost felt like a new...

www.inference.vc

2

46

254

Ferenc Huszár

@fhuszar

6 years

It's back-to-school time everyone! New post on "The Blessings of Multiple Causes" by @yixinwang_ and @blei_lab

4

74

252

Ferenc Huszár

@fhuszar

4 years

New massive recommender system dataset from Twitter! We hope this will stimulate more research on recommender systems which didn't really have datasets of this size.

Wenzhe Shi 🐕🐎

@trustswz

4 years

We are releasing the biggest ever (160 million samples) recommender system public dataset today with @recsyschallenge 2020. Please go check out the website:

7

110

328

4

45

251

Ferenc Huszár

@fhuszar

6 years

Here is a neat overview paper in which Judea Pearl outlines the specific tasks which one cannot solve with 'associational' reasoning and learning.

1

69

243

Ferenc Huszár

@fhuszar

3 years

Google develops AI to optimize the layout of its next generation AI chip. (a) human-designed layout (b) AI-designed layout is 30% more power-efficient, includes four legs, an on-chip battery and a syringe of 5G activated nanorobots.

6

32

235