Get Started for FREE
Sign up with Facebook Sign up with X
I don't have a Facebook or a X account
 Your new post is loading...
Your new post is loading...
Scoop.it!
From AlexNet to GPT-3, we curate a list of 10 papers that mark significant research advancements in machine learning, deep learning, computer vision, NLP, and reinforcement learning over the past 10 years. Author presentation and detailed paper reviews are also included. (Authored by Crossminds in Research Spotlights)
Juliette Decugis's insight:
Good resource to understand the 10 most cited AI papers (includes paper abstracts and descriptive videos, as well as links to the full papers).
As part of the UC Berkeley AI Research lab, almost every research paper I've encountered for the past 3 years relies on these 10 publications. Papers focused on improving the computational costs of NN use Transformers, AlexNet, GPT-3 and/or BERT as baseline models depending on the task. Most robustness papers rely on the methods developed by Goodfellow et al. in their GANs paper. Beyond research, all tech companies I've worked at derive their own deep learning models based on ensembles or specific applications of these models. 
No comment yet.
Sign up to comment
Scoop.it!
"Posted by Ehsan Amid, Research Scientist, and Rohan Anil, Principal Engineer, Google Research, Brain Team"
Juliette Decugis's insight:
Many recent ML papers hope to address the overwhelming problem of deep learning models: their computational and memory cost. Whereas lots of recent work has focused on the sparsification of said models, LocoProp attempts to rethink backpropagation - the most expensive step of neural network training.
LocoProp decomposes a model's objective function into a layer-wise loss, comparing a layer's output and the overall bath's final output, accompanied by a regularizer term (L2 loss). Breaking down the loss function across layers permits parallelization of training, smaller order calculations and more flexibility. Furthermore, the paper demonstrates that "the overall behavior of the combined updates closely resembles higher-order updates." Potential limits of the paper: "small" networks used, "still remains to be seen how well the method generally works across tasks." Link to the complete paper: https://proceedings.mlr.press/v151/amid22a/amid22a.pdf
Scoop.it!
In 2019, Amazon’s research team developed a deep learning method called DeepAR that exhibits a ~15% accuracy boost relative to state-of-the-art TS forecasting models. It’s robust out-of-the-box and…
Juliette Decugis's insight:
Amazon's DeepAR network combines traditional auto-regressive forecasting methods based on a time series' attributes (trend, seasonality...) and RNN. The model learns jointly from multiple related time series even with high differences in series' magnitudes.
Scoop.it!
Skin conditions affect 1.9 billion people. Because of a shortage of dermatologists, most cases are seen instead by general practitioners with lower diagnostic accuracy. We present a deep learning system (DLS) to provide a differential diagnosis of skin conditions using 16,114 de-identified cases (photographs and clinical data) from a teledermatology practice serving 17 sites. [...] Our DLS can potentially augment the current clinical workflow in a primary-care setting in several ways.
Juliette Decugis's insight:
Using the Tensorflow framework, Yuan Liu and Ayush Jain's team reached new steps in medical deep learning. Their deep learning system not only categorizes skin conditions as benign or malignant, it successfully identifies 1 out of 26 skin conditions and provides a complete diagnosis for each patient. Its results compare to those of a dermatologist. How can we explain this success? The DLS relies on multiple images per case as well as metadata (+4% accuracy observed) to make predictions. Furthermore, the system makes two level predictions based on the rarity of skin conditions. It's innovative in its data utility as well as system architecture.
Scoop.it!
Posted by Adam Roberts, Staff Software Engineer and Colin Raffel, Senior Research Scientist, Google Research Over the past few years
Juliette Decugis's insight:
Meet the new superstar of NLP: the T5 text to text model! T5 provides one unique model that can be applied to almost all existing NLP tasks: translation, summarization, classification, regression... It already provides state of the art results in many QA benchmark tasks.
Scoop.it!
From
medium
I n the world of machine learning, neural network and associated deep learning models are quickly becoming dominant, with very significant amounts of work being published every day, often…
Juliette Decugis's insight:
Implicit Neural Networks, sometimes called deep equilibrium models, aim to represent neural networks as feedforward recursive loops. They provide unique advantages such as higher expressive power, robustness, and interpretability (through their sensitivity matrices).
Scoop.it!
From
medium
This is a story about how I’ve published a cry for help on my Linkedin less than a week ago and somehow found myself coordinating a group of 350 PhDs and AI specialists that are working 24/7 without…
Juliette Decugis's insight:
Article by Artur Kiulian creator of CoronaWhy.org, a search engine powered by machine learning algorithms to help medical researchers navigate scholar articles related to COVID-19. Discover how AI can help us fight and cure coronavirus. Currently, the CoranaWhy search engine provides specific articles related to a given query, data visualization tools for locating the spread of the virus, a matrix for exploring current treatments...
Scoop.it!
Hyper-intelligent algorithms are not going to take over the world for these five reasons.
Juliette Decugis's insight:
Sorry robots will not rule the world! When we think of AI, we see a future with superhuman robots destroying our species. Hurray, that is not the case! This article shows with five simple points the impossibility of this scenario.
Scoop.it!
Artificial intelligence is being developed to analyze written text to block terror-related content.
Juliette Decugis's insight:
Facebook has been greatly criticised for not doing enough to stop the spread of terrorist ideas on its platform. After the recent London attacks, the social media decided to change its policy and spend more time solving this issue through the potential use of AI.
Scoop.it!
There is concern that cutting-edge technologies developed in the U.S. could be used by China to bolster its military capabilities and even push it ahead in strategic industries
Juliette Decugis's insight:
Military use of artificial intelligence is becoming more and more frequent especially in drone warfare. It is no surprise that China has developed a growing interest in Silicon Valley's AI companies. Each year, the country invests 40 million dollars in American companies. The investements are taking a negative toll on the US economy and posing a potential threat.
Scoop.it!
The age of artificial intelligence, or AI, is here. Learn which jobs it will replace and which jobs are safe ... for now.
Juliette Decugis's insight:
Some jobs remain irreplaceable. AI is developing more and more in the recent years, but we have no clue when the revolution will actually come to life. It could be in the next years or in the next decades.
Scoop.it!
From
www
The founders of Microsoft and Facebook have cautionary words for the Class of 2017: At his Harvard University commencement speech on Thursday, Facebook chief executive Zuckerberg had some tough words for the Class of 2017.
Juliette Decugis's insight:
AI will create 15 million new jobs in the next years but are also putting 38% of jobs in the U.S. at “high risk” of being replaced.
Scoop.it!
Traffic lights fitted with artificial intelligence could spell the end of rush hour queues in British cities.
Juliette Decugis's insight:
The beginning of smart cities. The British city wants to install AI powered cameras into the traffic lights in order to monitor traffic based on the number of cars and to prioritize ambulances, for example. COMING SOON
|
Scoop.it!
"I think we should think of AI as the intellectual equivalent of a backhoe. It will be much better than us at a lot of things" Artificial intelligence is a booming industry in 2019 with lots of new technological advancements.
Juliette Decugis's insight:
It blows my mind to think the most widely used deep learning models were developed a little less than 40 years ago. Transformers (2017) do seem to be missing on the deep learning timeline...
Scoop.it!
From
arxiv
"we propose Stochastic Activation Pruning (SAP), a mixed strategy for adversarial defense. SAP prunes a random subset of activations (preferentially pruning those with smaller magnitude) and scales up the survivors to compensate. [...] Experiments demonstrate that SAP confers robustness against attacks, increasing accuracy and preserving calibration..
Juliette Decugis's insight:
SAP uses a minimax zero-sum game from game theory between adversary and model to increase robustness. The activation maps are converted to a multinomial distribution and sampled during each forward pass based on their magnitudes. The model defines one adversary, meant to perturb the model, and one defender, meant to adapt the model's parameters based on the perturbations.
Unique advanatages: no retraining or fine tuning
Scoop.it!
People have a tendency to treat technology and data as neutral, sterile and immune to mortal failings. Yet the digital tools we use at schools, job
Juliette Decugis's insight:
Ruha Benjamin highlights the biases of current AI and the risks of seeing technology as neutral. In her title, "The New Jim Code" she references the 2010 book "The New Jim Crow" by Alexander Michelle. Artificial intelligence is creating a new illusion of a 'colorblind society' where people of color are rendered invisible to everyday technologies and hypervisible to policing.
Scoop.it!
Neural models in recent years have been successful in almost every field including extremely complex problem statements. However, these models are huge in size, with millions (and billions) of…
Juliette Decugis's insight:
A comprehensive introduction to knowledge distillation, a new way to compress neural networks. Neural networks fascinate the machine learning community however there is one issue: they require millions to billions of parameters. Imagine if networks could teach each other reduced versions of their information while achieving the same results.
Scoop.it!
From
blog
We are thrilled to have mastered Go and thus achieved one of the grand challenges of AI.
Juliette Decugis's insight:
In 2015, AlphaGo created by DeepMind became the first computer program to beat a professional player in the game of Go. Playing against itself, AlphaGo learned to master not only the most complex Go moves but more importantly discovered new techniques unknown to champions. AlphaGo could represent the start of the future of AI.
Scoop.it!
Researchers are revising the ImageNet data set. But algorithmic anti-bias training is harder than it seems.
Juliette Decugis's insight:
Many fields, such as judicial departments, are already starting to use AI as a way of making more neutral decisions. Many people view AI as a better version of humans capable of more informed judgments. But who do machines learn from? Humans.
Scoop.it!
Google's Geoff Hinton helped catalyze the current AI boom and says he knows how to make machines smarter at understanding the world.
Juliette Decugis's insight:
Do you want to know who first made AI possible? Geoff Hinton. What's his next move to simplify data learning? Read all the details in this well-summarized article.
Scoop.it!
81% of IT leaders are currently investing in or planning to invest in Artificial Intelligence (AI) and machine learning. Based on the study, CIOs have a new mandate to integrate AI into IT technology stacks. The study found that 43% are evaluating and doing a Proof of Concept (POC).
Juliette Decugis's insight:
Artificial Intelligence is already a key tool for software companies. This study shows how much it influences IT leaders and its growth in the last years.
Scoop.it!
A look at where Artificial Intelligence is heating up, from healthcare to entertainment. Via Farid Mheir
Farid Mheir's curator insight,
June 11, 2017 4:37 PM
WHY THIS IS IMPORTANT AI startups funding shows that the investments has been growing steadily in the past 4 years and that every industry will be affected by the technology.
Scoop.it!
Get ready for the future - artificial intelligence is here to stay, and help marketers understand their customer's needs, emotions and preferences better.
Juliette Decugis's insight:
How companies use artificial intelligence today: a strong marketing tool.
Scoop.it!
According to sources from TimesOfIndia, the Artificial Intelligence (AI) system, developed by researchers from University of Adelaide in Australia, analysed the medical imaging of 48 patients' by simply looking at the images of their organs and were able to predict which of them would die within five years, with 69 percent accuracy.
Juliette Decugis's insight:
Will AI replace even jobs requiring decades of higher education? Doctors are also at risk in this coming revolution...
Scoop.it!
Security researchers at MIT have developed an Artificial Intelligence-based cyber security platform that can predict, detect, and stop 85% of Cyber Attacks with high accuracy.
Juliette Decugis's insight:
"What if we could Predict when a cyber attack is going to occur before it actually happens and prevent it? Isn't it revolutionary idea for Internet Security?"
|
