Stay updated with breaking news from Rogerio Feris. Get real-time updates on events, politics, business, and more. Visit us for reliable news and exclusive interviews.
PhD students Athul Paul Jacob, Maohao Shen, Victor Butoi, and Andi Peng, interning with the MIT-IBM Watson AI Lab, look to improve natural language usage in AI models so that the AI systems can be more dependable and accurate.
Research finds using a large collection of simple, un-curated synthetic image generation programs to pretrain a computer vision model for image classification yields greater accuracy than employing other pretraining methods that are more costly and time consuming, and less scalable.
Machine-learning models trained to classify human actions using synthetic data can outperform models trained using real data in certain situations. This could help scientists identify when it’s better to use synthetic data for training, which could eliminate bias, privacy, security, and copyright issues that often impact real datasets.
The advances that emerging AI and ML models can bring are dependent on the speed at which we can train them. Synthetic data may be the key.
A machine learning model called VALHALLA uses a trained neural network to understand and digest a source sentence in one language, hallucinate an image of what that sentence describes, and translate into a target language. The work was led by researchers from MIT, IBM, and UC San Diego.