Articles for category: AI Research

March 9, 2025

ikayaniaamirshahzad@gmail.com

SCIN: A new resource for representative dermatology images

Posted by Pooja Rao, Research Scientist, Google Research Health datasets play a crucial role in research and medical education, but it can be challenging to create a dataset that represents the real world. For example, dermatology conditions are diverse in their appearance and severity and manifest differently across skin tones. Yet, existing dermatology image datasets often lack representation of everyday conditions (like rashes, allergies and infections) and skew towards lighter skin tones. Furthermore, race and ethnicity information is frequently missing, hindering our ability to assess disparities or create solutions. To address these limitations, we are releasing the Skin Condition Image

March 9, 2025

ikayaniaamirshahzad@gmail.com

Ideas: Quantum computing redefined with Chetan Nayak

CHETAN NAYAK: Thank you. Thanks for having me. And I’m excited to tell you about this stuff. HUIZINGA: Well, you have a huge list of accomplishments, accolades, and awards—little alliteration there. But I want to start by getting to know a bit more about you and what got you there. So specifically, what’s your “research origin story,” as it were? What big idea inspired you to study the smallest parts of the universe? NAYAK: It’s a great question. I think if I really have to go back to the origin story, it starts when I was a kid, you know,

March 9, 2025

ikayaniaamirshahzad@gmail.com

Weight Banding

This article is part of the Circuits thread, an experimental format collecting invited short articles and critical commentary delving into the inner workings of neural networks. Branch Specialization Introduction Open up any ImageNet conv net and look at the weights in the last layer. You’ll find a uniform spatial pattern to them, dramatically unlike anything we see elsewhere in the network. No individual weight is unusual, but the uniformity is so striking that when we first discovered it we thought it must be a bug. Just as different biological tissue types jump out as distinct under a microscope, the weights

March 8, 2025

ikayaniaamirshahzad@gmail.com

An Axiomatic Definition of Hierarchical Clustering

An Axiomatic Definition of Hierarchical Clustering Ery Arias-Castro, Elizabeth Coda; 26(10):1−26, 2025. Abstract In this paper, we take an axiomatic approach to defining a population hierarchical clustering for piecewise constant densities, and in a similar manner to Lebesgue integration, extend this definition to more general densities. When the density satisfies some mild conditions, e.g., when it has connected support, is continuous, and vanishes only at infinity, or when the connected components of the density satisfy these conditions, our axiomatic definition results in Hartigan’s definition of cluster tree. [abs] [pdf][bib]        Source link

March 8, 2025

ikayaniaamirshahzad@gmail.com

Improving lung cancer pathological hyperspectral diagnosis through cell-level annotation refinement

The framework of this study is illustrated in Fig. 1. Initially, we developed a custom hyperspectral microscopy system to acquire raw hyperspectral data from H&E-stained pathological tissue samples. Next, we performed a series of preprocessing steps and label preparation on the hyperspectral data to establish a high-quality hyperspectral dataset. Subsequently, we employed random forest (RF), SVM, 2D CNN, 3D CNN, and an improved hybrid CNN model to identify tumor tissues. Specifically, our improved model, named HybridSN-Att, draws inspiration from the recent work in remote sensing of Roy et al.30, which combines the 2D CNN and 3D CNN to enhance performance. Fig.

March 8, 2025

ikayaniaamirshahzad@gmail.com

RLHF and RLAIF in GPT-NeoX

Today SynthLabs and EleutherAI are excited to announce large scale post training and preference learning in GPT-NeoX, one of the most widespread and adopted pretraining frameworks for large scale language models. This effort represents a partnership towards improving accessibility of preference learning research at scale. Currently large scale preference learning research is bottlenecked by a lack of easily scalable and robust frameworks. Pushing the boundary of what models are easily trainable and what training methodologies are easily accessible will enable a new wave of research developments and breakthroughs in the space of preference learning as well as a new set

March 8, 2025

ikayaniaamirshahzad@gmail.com

Stanford CRFM

Your browser does not support the audio tag. Researchers teamed up with Carol Reiley and players from the San Francisco Symphony to perform music generated by the Anticipatory Music Transformer. Figure 1. A violin accompaniment of Für Elise, generated by an Anticipatory Music Transformer.Featured performers: In Sun Jang (violin) and John Wilson (piano). Generative models offer exciting new tools and processes for artistic work. This post describes the use of a generative model of music—the Anticipatory Music Transformer—to create a new musical experience. We used the model to compose a violin accompaniment to Beethoven’s Für Elise. On a technical level,

March 8, 2025

ikayaniaamirshahzad@gmail.com

Function Calling at the Edge – The Berkeley Artificial Intelligence Research Blog

The ability of LLMs to execute commands through plain language (e.g. English) has enabled agentic systems that can complete a user query by orchestrating the right set of tools (e.g. ToolFormer, Gorilla). This, along with the recent multi-modal efforts such as the GPT-4o or Gemini-1.5 model, has expanded the realm of possibilities with AI agents. While this is quite exciting, the large model size and computational requirements of these models often requires their inference to be performed on the cloud. This can create several challenges for their widespread adoption. First and foremost, uploading data such as video, audio, or text

March 8, 2025

ikayaniaamirshahzad@gmail.com

Updates to Veo, Imagen and VideoFX, plus introducing Whisk in Google Labs

While video models often “hallucinate” unwanted details — extra fingers or unexpected objects, for example — Veo 2 produces these less frequently, making outputs more realistic. Our commitment to safety and responsible development has guided Veo 2. We have been intentionally measured in growing Veo’s availability, so we can help identify, understand and improve the model’s quality and safety while slowly rolling it out via VideoFX, YouTube and Vertex AI. Just like the rest of our image and video generation models, Veo 2 outputs include an invisible SynthID watermark that helps identify them as AI-generated, helping reduce the chances of

March 8, 2025

ikayaniaamirshahzad@gmail.com

Apply now for Google for Startups Accelerator: AI for Energy

Today, various startups are using AI to shape the future of energy. They’re building products that enhance grid resiliency, maximize value of existing assets and accelerate analysis and planning tools that keep our system running. To support these AI-focused startups in scaling rapidly and responsibly, we’re introducing the Google for Startups Accelerator: AI for Energy (applications open today) with cohorts running this year in North America and Europe. This equity-free program offers 10 weeks of intensive mentorship and technical project support to startups integrating AI into their core energy services or products. Selected startups will collaborate with a cohort of