The company demonstrated its latest AI model using a small dataset – just a fraction of the size typically used for a Generative Adversarial Network (GAN) – of artwork from the Metropolitan Museum of Art.

From the dataset, NVIDIA’s AI was able to create new images which replicate the style of the original artist’s work. These images can then be used to help train further AI models.

The AI achieved this impressive feat by applying a breakthrough neural network training technique similar to the popular NVIDIA StyleGAN2 model. 

The technique is called Adaptive Discriminator Augmentation (ADA) and NVIDIA claims that it reduces the number of training images required by 10-20x while still getting great results.

David Luebke, VP of Graphics Research at NVIDIA, said:

“These results mean people can use GANs to tackle problems where vast quantities of data are too time-consuming or difficult to obtain.

I can’t wait to see what artists, medical experts and researchers use it for.”

Healthcare is a particularly exciting field where NVIDIA’s research could be applied. For example, it could help to create cancer histology images to train other AI models.

The breakthrough will help with the issues around most current datasets.

Large datasets are often required for AI training but aren’t always available. On the other hand, large datasets are difficult to ensure their content is suitable and does not unintentionally lead to algorithmic bias.

Earlier this year, MIT was forced to remove a large dataset called 80 Million Tiny Images. The dataset is popular for training AIs but was found to contain images labelled with racist, misogynistic, and other unacceptable terms.

A statement on MIT’s website claims it was unaware of the offensive labels and they were “a consequence of the automated data collection procedure that relied on nouns from WordNet.”

The statement goes on to explain the 80 million images contained in the dataset – with sizes of just 32×32 pixels – meant that manual inspection would be almost impossible and couldn’t guarantee all offensive images would be removed.

By starting with a small dataset that can be feasibly checked manually, a technique like NVIDIA’s ADA could be used to create new images which emulate the originals and can scale up to the required size for training AI models.

In a blog post, NVIDIA wrote:

“It typically takes 50,000 to 100,000 training images to train a high-quality GAN. But in many cases, researchers simply don’t have tens or hundreds of thousands of sample images at their disposal.

With just a couple thousand images for training, many GANs would falter at producing realistic results. This problem, called overfitting, occurs when the discriminator simply memorizes the training images and fails to provide useful feedback to the generator.”

You can find NVIDIA’s full research paper here (PDF). The paper is being presented at this year’s NeurIPS conference as one of a record 28 NVIDIA Research papers accepted to the prestigious conference.

The post NVIDIA breakthrough emulates images from small datasets for groundbreaking AI training appeared first on AI News.

In the name of research, here’s one I made earlier:

Doublicat uses a Generative Adversarial Network (GAN) to do its magic. The GAN is called RefaceAI and is developed by a company of the same name.

RefaceAI was previously used in a face swapping app called Reflect. Elon Musk once used Reflect to put his face on Dwayne Johnson’s body. 

The app is a lot of fun, but – after concerns about viral Russian app FaceApp – many will be wondering just how much data is being collected in return.

Doublicat’s developers are upfront with asking for consent to store your photos upon first opening the app and this is confirmed in their privacy policy: “We may collect the photos, that you take with your camera while using our application.”

However, Doublicat says that photos are only stored on their server for 24 hours before they’re deleted. “The rest of the time your photos used in Doublicat application are stored locally on your mobile device and may be removed any time by either deleting these photos from your mobile device’s file system.”

The app also collects data about facial features but only the vector representations of each person’s face is stored. Doublicat assures that the facial recognition data collected “is not biometric data” and is deleted from their servers within 30 calendar days.

“In no way will Doublicat use your uploaded content for face recognition as Doublicat does not introduce the face recognition technologies or other technical means for processing biometric data for the unique identification or authentication of a user.”

The amount of data Doublicat can collect is limited compared to some alternatives. Apps such as Zao require users to 3D model their face whereas Doublicat only takes a front-facing picture.

RefaceAI is now looking to launch an app which can make deepfake videos rather than just GIFs. The move is likely to be controversial given the concerns around deepfakes and how such videos could be used for things such as political manipulation.

A fake video of Nancy Pelosi, Speaker of the United States House of Representatives, went viral last year after purportedly showing her slurring her words as if she was intoxicated. The clip shows how even a relatively unsophisticated video (it wasn’t an actual deepfake in this case) could be used to cause reputational damage and even swing votes.

A report from the NYU Stern Center for Business and Human Rights last September, covered by our sister publication MarketingTech, highlighted the various ways disinformation could be used ahead of this year’s presidential elections. One of the eight predictions is that deepfake videos will be used “to portray candidates saying and doing things they never said or did”.

Earlier this month, Facebook announced new policies around deepfakes. Any deepfake video that is designed to be misleading will be banned. The problem with the rules is they don’t cover videos altered for parody or those edited “solely to omit or change the order of words,” which will not sound encouraging to anyone wanting a firm stance against manipulation.

Doublicat is available for Android and iOS.

Interested in hearing industry leaders discuss subjects like this? , , , AI &

The post Deepfake app puts your face on GIFs while limiting data collection appeared first on AI News.

During a midnight keynote, Amazon unveiled Transcribe Medical, SageMaker Operators for Kubernetes, and DeepComposer.

Transcribe Medical

The first announcement we’ll be talking about is likely to have the biggest impact on people’s lives soonest.

Transcribe Medical is designed to transcribe medical speech for primary care. The feature is aware of medical speech in addition to standard conversational diction.

Amazon says Transcribe Medical can be deployed across “thousands” of healthcare facilities to provide clinicians with secure note-taking abilities.

Transcribe Medical offers an API and can work with most microphone-equipped smart devices. The service is fully managed and sends back a stream of text in real-time.

Furthermore, and most importantly, Transcribe Medical is covered under AWS’ HIPAA eligibility and business associate addendum (BAA). This means that any customer that enters into a BAA with AWS can use Transcribe Medical to process and store personal health information legally.

SoundLines and Amgen are two partners which Amazon says are already using Transcribe Medical.

Vadim Khazan, president of technology at SoundLines, said in a statement:

“For the 3,500 health care partners relying on our care team optimisation strategies for the past 15 years, we’ve significantly decreased the time and effort required to get to insightful data.”

SageMaker Operators for Kubernetes

The next announcement is Amazon SageMaker Operators for Kubernetes.

Amazon’s SageMaker is a machine learning development platform and this new feature lets data scientists using Kubernetes train, tune, and deploy AI models.

SageMaker Operators can be installed on Kubernetes clusters and jobs can be created using Amazon’s machine learning platform through the Kubernetes API and command line tools.

In a blog post, AWS deep learning senior product manager Aditya Bindal wrote:

“Customers are now spared all the heavy lifting of integrating their Amazon SageMaker and Kubernetes workflows. Starting today, customers using Kubernetes can make a simple call to Amazon SageMaker, a modular and fully-managed service that makes it easier to build, train, and deploy machine learning (ML) models at scale.”

Amazon says that compute resources are pre-configured and optimised, only provisioned when requested, scaled as needed, and shut down automatically when jobs complete.

SageMaker Operators for Kubernetes is generally available in AWS server regions including US East (Ohio), US East (N. Virginia), US West (Oregon), and EU (Ireland).

DeepComposer

Finally, we have DeepComposer. This one is a bit more fun for those who enjoy playing with hardware toys.

Amazon calls DeepComposer the “world’s first” machine learning-enabled musical keyboard. The keyboard features 32-keys and two octaves, and is designed for developers to experiment with pretrained or custom AI models.

In a blog post, AWS AI and machine learning evangelist Julien Simon explains how DeepComposer taps a Generative Adversarial Network (GAN) to fill in gaps in songs.

After recording a short tune, a model for the composer’s favourite genre is selected in addition to setting the model’s parameters. Hyperparameters are then set along with a validation sample.

Once this process is complete, DeepComposer then generates a composition which can be played in the AWS console or even shared to SoundCloud (then it’s really just a waiting game for a call from Jay-Z).

Developers itching to get started with DeepComposer can apply for a physical keyboard for when they become available, or get started now with a virtual keyboard in the AWS console.

The post Amazon makes three major AI announcements during re:Invent 2019 appeared first on AI News.