Discover and read the best of Twitter Threads about #deeplearning

Most recents (24)

A paper on "AlphaFold-multimer", a version of AlphaFold that works on protein complexes, was released by @DeepMind.

Accurately predicted structures can lead to better understand the function of such protein complexes that underpin many biological processes!

#DeepLearning 1/4
Before "AlphaFold-multimer", people discovered that AlphaFold can predict complexes if you connect them with a long linker (this tweet was cited in the above paper!) 2/4
The new model, which had various adjustments to handle the larger protein complex structures, shows improved performance over this linker approach, along with other approaches 3/4
Read 4 tweets
1/7 #SeaTurtleTalks
Automatic identification of immature green turtle behavior from accelerometer reveals key marine conservation area in the Caribbean

Accelerometer records animal movement & body posture. Signal interpretation can be difficult & limits its use on #seaturtle
2/7 We deployed multi-sensor recorders (accelerometer, gyroscope and time-depth recorder) combined with animal-borne video-recorder to validate the signal on immature green turtles in Martinique.
3/7 From labelled signals, we trained a #deeplearning algorithm, #Vnet, to automatically identify green turtle #behavior.

🚩High ability to detect behaviors with low discriminative signals such as Feeding and Scratching, and precision down to one centi-second. #rblt
Read 7 tweets
Surprised #deeplearning methods were not massively used to detect Covid? Some tried, but most failed: e.g. they rely on patients' age instead of 'truly' analyzing the medical scans. This is because networks learn correlation and not causation. Our preprint Fishr tackles this. 1/4
When dealing with multiple datasets, we intuited that the learning should unfold consistently across those: i.e., the gradient distributions should be independent of the dataset. After a few approximations, we simply bring closer the dataset-level gradient variances! 2/4
Our approach is closely related to the Fisher Information and the Hessian. This explains the name of our work - Fishr - and why this aligns the loss landscapes per dataset. This works beyond expectations, outperforming all previous methods on the public DomainBed benchmark 3/4
Read 4 tweets
"torch.manual_seed(3407) is all you need"!
draft 📜: davidpicard.github.io/pdf/lucky_seed…
Sorry for the title. I promise it's not (entirely) just for trolling. It's my little spare time project of this summer to investigate unaccounted randomness in #ComputerVision and #DeepLearning.
🧵👇 1/n
The idea is simple: after years of reviewing deep learning stuff, I am frustrated of never seeing a paragraph that shows how robust the results are w.r.t the randomness (initial weights, batch composition, etc). 2/n
After seeing several videos by @skdh about how experimental physics claims tend to disappear through repetition, I got the idea of gauging the influence of randomness by scanning a large amount of seeds. 3/n
Read 11 tweets
1/

Thread of the very best #YouTube channels and #Twitter accounts to follow for:

#AI/ #ML, #DeepLearning, #neural and all things #datascience

bit.ly/3g8pVDL

#AI #machinelearning @wiserin10 #datascience #bigdata #artificialintelligence
2/

@Analyticsindiam

Analytics India Magazine includes discussions on news, tips for the data ecosystem and a deep dive into #AI/#ML, #deeplearning and #neural networks

#YouTube subscriber count: 38k
3/

@Krishnaik06

Krish Naik is co-founder of iNeuron.ai and specialises in #machinelearning, #deeplearning, and computer vision. Krish’s #YouTube channel is a deep dive into all things #AI/#ML, perfect for beginners

YouTube subscriber count: 421k
Read 15 tweets
1/ Interested in how Deep Learning and AI is impacting a 50-year old grand challenge in biology (protein structure folding)?
See this thread 👀🧵👇
#deeplearning #AI #biology #bioinformatics
2/ Deepmind's Alphafold2 Solves Protein Structures (Part 1) #shorts
3/ Deepmind's Alphafold2 Solves Protein Structures (Part 2) #shorts
Read 5 tweets
I've written multiple threads on how to get started with #MachineLearning, #DeepLearning , and #DataScience in general.
check them out (Bookmark).
🧵👇👇
Read 5 tweets
Had a great time attending @Geccoconf (virtually) this week. Thought I would dump some opinions/reflections into the twitter bit bucket before I forget them (thread)
1/ The organizers did an outstanding job of making this conference work well online. Talks on zoom, posters and social in gather.town. Sure, it's not as good as in person and time zones are a pain, but the low cost and accessibility are big positives.
2/ I was very impressed by the quality of the papers and talks. In the past there were times when some of the #machinelearning community would have scoffed at #evolutionarycomputation for being full of weird heuristics and crappy experimental standards.
Read 7 tweets
Happy to announce that our (@zhao7900, @YinyingWang) #deeplearning method INSCT for integration of #scrnaseq is out @NatMachIntell go.nature.com/2Uq73If! We extended triplet neural nets by defining 'batch-aware' triplet sampling, which overcomes batch effects in #scrnaseq data.
Instead of sampling triplets the traditional way, we added constraint to sample anchor-positives from different & anchor-negatives from the same batch. (We expect this approach to be also useful outside of #scrnaseq data - happy to discuss w anyone interested).
The semi-supervised mode allows robust transfer of cell labels across batches. Use case example: transfer labels from reference to new data set.
Read 4 tweets
Daily Bookmarks to GAVNet 06/09/2021 greeneracresvaluenetwork.wordpress.com/2021/06/09/dai…
China’s Hot Summer Is Latest Test of Its Carbon-Neutrality Drive

bloomberg.com/news/articles/…

#china #ClimateChange #CarbonNeutrality #consequences
Quantum Computing and Reinforcement Learning Are Joining Forces to Make Faster AI

singularityhub.com/2021/03/16/qua…

#QuantumComputing #ReinforcementLearning #ArtificialIntelligence
Read 10 tweets
We may have found a solid hypothesis to explain why extreme overparametrization is so helpful in #DeepLearning, especially if one is concerned about adversarial robustness. arxiv.org/abs/2105.12806
1/7
With my student extraordinaire Mark Sellke @geoishard, we prove a vast generalization of our conjectured law of robustness from last summer, that there is an inherent tradeoff between # neurons and smoothness of the network (see *pre-solution* video). 2/7
If you squint hard enough (eg, like a physicist) our new universal law of robustness even makes concrete predictions for real data. For ex. we predict that on ImageNet you need at least 100 billion parameters (i.e., GPT-3-like scale) to possibly attain good robust guarantees. 3/7
Read 7 tweets
There are way too many papers on #MachineLearning and on #DeepLearning these days. How to choose which papers to read? A tiny thread 🧵
The first one is our absolute favorite. Arxiv sanity by none other than @karpathy!

Link: arxiv-sanity.com
The second one is by @labmlai. It is pretty new and the interface is pretty smooth!

Link: papers.labml.ai/papers/recent/
Read 6 tweets
"Attention is all you need" is one of the most cited papers in last couple of years. What is attention? Let's try to understand in this thread.

Paper link: arxiv.org/abs/1706.03762

#DeepLearning #MachineLearning #Transformers
In Self-attention mechanism, we are updating the features of a given point, with respect to other features. The attention proposed in this paper is also known as Scaled dot-product attention.
Lets say, our data point is a single sentence, we embed each word into some d-dimensional space, so we compute how each point is similar to each other point, and weigh its representation accordingly. The similarity matrix is just a scaled dot product!
Read 7 tweets
Here are my experiences learning about RL over the past 3 months! ♥️ Hopefully, this will be the most beginner-friendly guide out there.

blog: gordicaleksa.medium.com/how-to-get-sta…

@DeepMind @OpenAI Image
I've tried to give you all of the tips and tricks I could think of both for more productive learning in general and stuff specific to the RL field.
The structure of the blog:
1) Intro
2) RL 101 (getting you exposed to the terminology)
3) Cool things about RL (awesome RL apps!)
4) RL is not just roses
5) Getting started with RL
6) Going deeper - reading papers
7) Implementing an RL project from scratch
8) Related subfields
Read 4 tweets
Worked with @andrecnferreira on a end to end #DeepLearning project for @full_stack_dl. I learned a lot from fundamentals of DL to #testing, #troubleshooting #deployment and #AIEthics. Thanks for organising #FSDL course @sergeykarayev @josh_tobin_ @pabbeel
Trained a @fastdotai model
Set up data infra on @googlecloud
Developed + deployed a @streamlit dashboard

Github Code: github.com/karthikraja95/…
Read 4 tweets
I just open-sourced my implementation of the original @DeepMind's DQN paper! But this time it's a bit different!

There are 2 reasons for this, see the thread.

GitHub: github.com/gordicaleksa/p…

#rl #deeplearning
1) This time the project is still not completely ready**. I'm yet to achieve the published results - so I encourage you to contribute!

Many of you have been asking me whether you can work on a project with me and I'll finally start doing it that way - from now onwards. ❤
2) This repo has the ambition to grow and become the go-to resource for learning RL. So collaborators are definitely welcome as I won't always have the time myself.

** main reasons are:
a) I was very busy over the last 2 weeks
b) It currently takes ~5 days to fully train DQN
Read 7 tweets
Deep dive into "ZeRO: Memory Optimizations Toward Training Trillion Parameter Models" by Samyam Rajbhandari, Olatunji Ruwase, Yuxiong He & @jeffra45

It proposes an optimizer to build huge language pre-trained models.

Thread👇🏼 🔎
thesequence.substack.com/p/-edge22-mach…
Zero Redundancy Optimizer (ZeRO) is an optimization module that maximizes both memory and scaling efficiency.

2/
It tries to address the limitations of data parallelism and model parallelism while achieving the merits of both

thesequence.substack.com/p/-edge22-mach…

3/
Read 7 tweets
ڈپریشن کو معمولی مت لیں
تحقیق بتاتی ہے کہ شدید ڈپریشن آگے چل کر ذہنی حالت کو اس قدر بگاڑ دیتا ہے انسان میں درد دینے اور درد سہنے کی صلاحیت ختم ہوجاتی ہے
ایسے ہی اک ماں نے اپنے بچوں کو آگ لگادی انہیں ذندہ جلتا دیکھ کر وہ سگریٹ پیتی رہی جب وہ نارمل ہوئی تو اسکے بچے جل کر مرچکے تھے
پھر جب اس طرح کے واقعے ہوجاتے ہیں تو ہم بڑی بڑی فلاسفی جھاڑنا شروع کر دیتے ہیں کہ ماں تھی یا فرعون ایسا کیسے کرگئی؟
تو خدارا لوگوں کو خوش رہنے دیں کسی کی ذہنی بربادی کا سبب مت بنیں کوئی ناچ رہا ہے ناچنے دیں
ہر کسی کی ذندگی میں دخل اندازی کرنا بند کریں
کسی کو ڈپریشن میں پائیں
تو اسے ٹکے ٹکے کی باتیں سنانے کی بجائے اسے ہنسانے کی کوشش کریں
مگر نہیں ہمیں تو شوق ہے لوگوں میں گھس کے انہیں زلیل کرنے کا
جب تک کوئی پھندہ ڈال کر جھول نہیں مرتا
تب تک ہم مانتے ہی نہیں کے اگلا ڈپریشن میں تھا ؟؟؟

#DeepLearning
#depression
Read 4 tweets
Perguruan Tinggi dan Narasi Keilmuan di Media Sosial

Slide ini saya presentasikan di acara soft-launching Website baru @univ_indonesia. Di depan civitas academica UI tadi pagi.

Bonus: analisis ttg BRIN dan Babi Ngepet di bagian akhir slides (spt saya janjikan minggu lalu).😅

>
Soft Launching website baru UI.
PERTANYAAN

1/ Perlu kah perguruan tinggi dan civitas academica aktif membangun jejaring dan narasi tentang ilmu pengetahuan yang menjadi fokusnya di media sosial?
Read 39 tweets
We are pleased to announce that our #ZeroCostDL4Mic work is now published!
nature.com/articles/s4146…
And our @github
github.com/HenriquesLab/Z…
#DeepLearning and #Microscopy for all!
For the occasion, I’ve put a small Tweetorial together to explain what this is all about!
Thread 0/7 Image
Thread 1/7

You can implement Deep Learning for microscopy in several ways:
1- Using local resources (fast GPUs)
2- Using cloud-computing
3- Using pre-trained models

We think that (1) is uncommon, and that (3) can be unreliable. So we do (2)! Image
Thread 2/7

We built a platform using @GoogleColab and @ProjectJupyter notebooks that can perform training, quality control and prediction, all on the cloud and for free!

All of this dedicated to #microscopy ! Image
Read 12 tweets
1/ My notes from @scale_AI Transform✍️It covers:
- Building good data
- Future of ML frameworks
- Challenges for scalable deployment
- How to assess ML maturity

Thanks @alexandr_wang and team for organizing the best AI conference of 2021 thus far. Enjoy!

jameskle.com/writes/scale-t…
2/ @AndrewYNg: "The single most important thing that an MLOps team needs to do is to ensure consistently high-quality data throughout all stages of the ML project life cycle." Image
3/ @fchollet
The 4 #DeepLearning Trends That Keras Will Support:
- An Ever-Growing Ecosystem of Reusable Parts
- Increasing Automation
- Large-Scale Workflows In The Cloud
- Real-World Deployment ImageImage
Read 13 tweets
Today we will summarize Vision Transformer (ViT) from Google. Inspired by BERT, they have implemented the same architecture for image classification tasks.

Link: arxiv.org/abs/2010.11929
Code: github.com/google-researc…

#MachineLearning #DeepLearning
The authors have taken the Bert architecture and applied it on an images with minimal changes.Since the compute increases with the length of the sequence, instead of taking each pixel as a word, they propose to split the image into some ’N’ patches and take each of them as token.
So first take each patch, flatten it (which will be of length P²C), and project it linearly to dimension D. And in the 0th position add a ‘D’ dimensional embedding which will be learnt. Add positional encoding to these embedding.
Read 9 tweets

Related hashtags

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3.00/month or $30.00/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!