All articles tagged with #google research
Google has released a complete map of the male fruit fly brain, enabling researchers to simulate neural activity in video games. Engineers have successfully trained this digital model to play Doom and Beat Saber, with performance varying based on simulated arousal levels. This milestone follows the release of the female fly brain map and aims to advance neuroscience research.
Google's latest research indicates that quantum computers could break RSA encryption 20 times faster than previously thought, raising concerns about the security of Bitcoin and other cryptographic systems. While current quantum computers are far from capable of such tasks, the findings accelerate the timeline for potential quantum threats, prompting the crypto community to develop quantum-resistant solutions. A Bitcoin bounty has been launched to test quantum vulnerabilities, highlighting the urgency of preparing for a post-quantum cryptographic era.
Google Research has unveiled Lumiere, an AI video generation model that can create highly realistic and seamless videos from text prompts and images, addressing the challenge of generating coherent motion. The model's Space-Time U-Net architecture allows for smooth and hyper-realistic video generation, surpassing existing models in visual quality and motion. Lumiere can also be used for video editing, stylized generation, cinemagraphs, and inpainting. While not yet available to the public, demos of the model's capabilities can be viewed on the Lumiere website.
Gmail has implemented its "largest defense upgrades in recent years" to enhance spam detection by 38%. The improvements are attributed to the use of RETVec (Resilient & Efficient Text Vectorizer), an open-source approach developed by Google Research. RETVec allows models to achieve state-of-the-art classification performance while reducing computational costs, making it suitable for various use cases. Gmail's implementation of RETVec has not only improved spam detection but also reduced false positive rates and Tensor Processing Unit usage. Google has extensively tested RETVec and found it highly effective for security and anti-abuse applications.
Google researchers proposed a method that combines a standard autoencoder with a diffusion process to efficiently compress high-quality images using score-based generative models. The proposed method outperforms several recent generative approaches in terms of image quality and preserves details much better compared to state-of-the-art approaches. The study revealed specific details that can be useful for future research in this domain, such as the impact of noise schedule and the amount of noise injected during image generation.
Alphabet is merging its internal Google Research team called Brain with DeepMind to bring two groups focused on artificial intelligence closer together. Jeff Dean, who currently leads Google's AI efforts, will be promoted and given the title of chief scientist at Google, reporting to Pichai. DeepMind CEO Demis Hassabis will lead the development of "the most capable and responsible general AI systems." The move marks Google's latest reorganization in response to the rapid developments in AI.