Google Research
·
22 hours ago
Researchers studied how diffusion models generate novel images and data beyond their training examples, finding that the models' creativity stems from neural network regularization effects that blur the learned score function. In a one-dimensional example with two training points at +1 and -1, weight decay causes the neural network to learn a smoother approximation of the score function, allowing generated particles to settle between the training points rather than copying them exactly. This score smoothing mechanism enables diffusion models to balance realism with novelty by interpolating between training data points while maintaining the quality of high-dimensional data manifolds.
Apple ML Research
·
1 day ago
Researchers addressed uncertainty quantification for LLM function-calling, which enables large language models to use external tools autonomously. The study focuses on measuring model confidence before executing function calls that could have irreversible consequences like financial transfers or data deletion. Better uncertainty estimates allow systems to defer high-risk decisions to human operators or alternative fallback mechanisms.
Apple ML Research
·
1 day ago
CLaRa is a framework that improves retrieval-augmented generation by compressing documents and jointly optimizing retrieval and generation in a shared continuous space. The method achieves state-of-the-art performance on QA benchmarks at a compression rate of 16, meaning documents are reduced to 1/16th their original length while maintaining answer quality. This unified approach allows gradients to flow through both retrieval and generation modules, aligning document relevance with answer quality during training.
Apple ML Research
·
1 day ago
Researchers propose FAE, a framework that adapts pre-trained visual encoders for image generation by using a single attention layer to convert high-dimensional features into low-dimensional latents suitable for generative models. On ImageNet 256×256, FAE achieves an FID score of 1.48 without classifier-free guidance after 800 epochs and 2.08 after 80 epochs. The approach enables simpler adaptation of pre-trained representations across different generative model families including diffusion models and normalizing flows.
