Both IBM and Google recently advanced development of Text-to-Speech (TTS) systems to create high-quality digital speech. OpenAI found that, since 2012, the compute power needed to train TTS models has exploded to more than 300,000 times. IBM created a much less compute-intensive model for speech synthesis, stating that it is able to do so in real-time and adapt to new speaking styles with little data. Google and Imperial College London created a generative adversarial network (GAN) to create high-quality synthetic speech. Continue reading Google and IBM Create Advanced Text-to-Speech Systems

Researchers recently introduced a series of rigorous benchmark tasks that measure the performance of sophisticated language-understanding AI. Facebook AI Research with Google’s DeepMind, University of Washington and New York University introduced SuperGLUE last week, based on the idea that deep learning models for today’s conversational AI require greater challenges. SuperGLUE, which uses Google’s BERT representational model as a performance baseline, follows the 2018 introduction of GLUE (General Language Understanding Evaluation), and encourages the creation of models that can understand more nuanced, complex language. Continue reading SuperGLUE Is Benchmark For Language-Understanding AI

Social networks, dating services, photo websites and surveillance cameras are just some of the sources of a growing number of databases compiling people’s faces. According to privacy advocates, Microsoft and Stanford University are among the many groups gathering images, with one such repository holding two million images. All these photos will be used to allow neural networks to build pattern recognition, in the quest to create cutting edge facial recognition platforms. Some companies have collected images for 10+ years. Continue reading Privacy Concerns Grow Over Facial Recognition Data Sets

Google research scientists report that they have produced realistic frames from open source video data sets at scale. Neural networks are able to generate complete videos from only a start and end frame, but it’s the complexity, information density and randomness of video that have made it too challenging to create such realistic clips at scale. The scientists wrote that, to their knowledge, “this is the first promising application of video-generation models to videos of this complexity.” The systems are based on a neural architecture known as Transformers, as described in a Google Brain paper, and are autoregressive, “meaning they generate videos pixel by pixel.” Continue reading Google Scientists Generate Realistic Videos at Scale with AI

Facebook is bringing back FMV (full motion video) games, which use pre-recorded video files to display action. With the work of Facebook AI Research scientists, the new FMV games are much improved, with a system that can extract controllable characters from real-world videos and then control their motion, thus generating new image sequences. Facebook AI Research scientists, in collaboration with Tel Aviv University, also unveiled a system that, unsupervised, converts audio of one singer to the voice of another. Continue reading Facebook Uses AI to Improve Games, Swap Singers’ Voices

Disney Research and Rutgers University scientists just created an end-to-end model using artificial intelligence to produce a storyboard and video featuring text from movie screenplays. This kind of text-to-animation model is not new, but this research advances the state-of-the-art by producing animations without annotated data or pre-training. The researchers wrote that the system is “capable of handling complex sentences” and is intended to make creatives’ work “more efficient and less tedious.” Continue reading Disney, Rutgers Scientists Use AI to Generate Storyboards

At the NAB 2019’s Broadcast Cybersecurity course, Emblematic Group founder/chief executive Nonny de la Peña introduced deep fakes by showing pairs of images and asking the audience to call out which was fake and which was real. From paired images of Presidents Obama and Trump, among others, audience members were consistently unable to pick the correct “fake” image. University of Washington researchers created a very convincing — but fake — video of Barack Obama, she revealed, by using neural network AI and 14 hours of Obama footage. Continue reading Another Perspective on Deep Fakes: Threat and Opportunity

The Association for Computing Machinery (ACM) honored three artificial intelligence pioneers with the prestigious Turing Award for their work on neural networks. The Turing Award, often dubbed the Nobel Prize of computing, was launched in 1966 and includes a $1 million prize that the three honorees will share. Doctors Geoffrey Hinton, Yann LeCun and Yoshua Bengio helped lay the foundation for the evolution of technologies including facial recognition, digital assistants and self-driving vehicles. Continue reading AI Pioneers Are Honored with the Prestigious Turing Award

Nvidia made a number of compelling announcements at this week’s GPU Technology Conference (GTC 2019) in San Jose, California. The company unveiled its GauGAN AI image creator that uses generative adversarial networks (GANs) to turn sketches into nearly photorealistic images. As part of its cloud pursuits, the company unveiled its latest RTX server configuration that is designed for Hollywood studios and those who want to create visual content quickly (each server pod can support up to 1,280 GPUs). Nvidia also announced partnerships with 3D software makers including Adobe, Autodesk and Unity to integrate Nvidia’s RTX ray-tracing platform. Continue reading Nvidia Demos New Products at Deep Learning & AI Confab

Google has unveiled GPipe, an open-sourced library that makes training deep neural networks more efficient under the TensorFlow framework Lingvo for sequence modeling. According to Google AI software engineer Yanping Huang, “in GPipe … we demonstrate the use of pipeline parallelism to scale up DNN training,” noting that larger DNN models “lead to better task performance.” Huang and his colleagues published a paper on “Efficient Training of Giant Neural Networks Using Pipeline Parallelism.” Continue reading Google GPipe Library Speeds Deep Neural Network Training

Neuroscientist and Intel’s chief technology officer of AI Amir Khosrowshahi revealed that he is remaining at Intel with a team of researchers building an innovative integrated circuit (IC). The IC under development will feature transistors that will, hope the researchers, function at voltages as low as 100 millivolts, a step towards matching voltage of communication in the brain. The existence of such an IC would unleash power-hungry AI applications targeting climate change, waste management and other global problems. Continue reading Intel Team Focuses on Low Voltage Transistor to Power AI

Faye Francy, executive director of the Automotive Information Sharing and Analysis Center (Auto-ISAC), led a conversation about the impact of machine learning, deep learning and AI on the autonomous vehicle (AV) ecosystem. “They work together to bring great things — and possibly nefarious things — to the auto industry,” she said. Inivision AI chairman Seamus Hatch noted that the three terms aren’t interchangeable. “We’re many years behind the singularity,” he said. “It’s a machine trained to solve a specific problem faster and more accurately than a human.” Continue reading CES Panel: Impact of Evolving Tech on Autonomous Vehicles

Cambridge Consultants has trained artificial intelligence to reconstruct images from a highly distorted image or video feed in real time that is the AI’s best guess of what the source image originally looked like. The unique approach of their DeepRay product involves recognizing and correcting for a wide universe of distortion patterns. Cambridge Consultants’ first clients are self-driving car companies concerned with accurate object and risk detection, but DeepRay could also have a number of entertainment industry technical and story applications. Continue reading DeepRay Uses AI to Reconstruct Distorted Video in Real Time

DeepMind recently released the full evaluation of AlphaZero, a single system capable of playing “Go,” chess, and shogi (Japanese chess). This new project builds on AlphaGo, a program that beat one of the best players in the world at the board game “Go” in 2016, and AlphaGo Zero, software capable of mastering the game from first principles. AlphaZero represents a dramatic step forward in AI research as it is one of the first intelligent systems capable of generalizing solutions to new problems with little to no human input. Continue reading DeepMind’s Learning Algorithm Could Prove a Game-Changer

Intel revealed that it has made progress in an anonymized, encrypted method of model training. Industries such as healthcare that need a way to use AI tools on sensitive, personally identifiable information have been waiting for just such a capability. At the NeurIPS 2018 conference in Montreal, Intel showed off its open-sourced HE-Transformer that works as a backend to its nGraph neural network compiler, allowing AI models to work on encrypted data. HE-Transformer is also based on a Microsoft Research encryption library. Continue reading Intel Describes Tool to Train AI Models with Encrypted Data