A5044698998- Privacy-Preserving Technologies in Data
- Topic Modeling
- Neural dynamics and brain function
- Stochastic Gradient Optimization Techniques
- Artificial Intelligence in Healthcare and Education
- stochastic dynamics and bifurcation
- Web Data Mining and Analysis
- Machine Learning in Healthcare
- Advanced Memory and Neural Computing
- Explainable Artificial Intelligence (XAI)
- Human Mobility and Location-Based Analysis
- Natural Language Processing Techniques
- Neural Networks and Applications
- Law, AI, and Intellectual Property
- Neuroscience and Neural Engineering
- Adversarial Robustness in Machine Learning
- Advanced Neural Network Applications
- Digital Rights Management and Security
- EEG and Brain-Computer Interfaces
- Ethics in Business and Education
- Neuroscience, Education and Cognitive Function
- Internet Traffic Analysis and Secure E-voting
- Recommender Systems and Techniques
- Digital Accessibility for Disabilities
- Neuroethics, Human Enhancement, Biomedical Innovations
Google (United States)
2015-2025
Microsoft (United States)
2012
Princeton University
2000-2007
Princeton Public Schools
2006
Modern mobile devices have access to a wealth of data suitable for learning models, which in turn can greatly improve the user experience on device. For example, language models speech recognition and text entry, image automatically select good photos. However, this rich is often privacy sensitive, large quantity, or both, may preclude logging center training there using conventional approaches. We advocate an alternative that leaves distributed devices, learns shared model by aggregating...
Abstract Large language models (LLMs) have demonstrated impressive capabilities, but the bar for clinical applications is high. Attempts to assess knowledge of typically rely on automated evaluations based limited benchmarks. Here, address these limitations, we present MultiMedQA, a benchmark combining six existing medical question answering datasets spanning professional medicine, research and consumer queries new dataset questions searched online, HealthSearchQA. We propose human...
Recent artificial intelligence (AI) systems have reached milestones in "grand challenges" ranging from Go to protein-folding. The capability retrieve medical knowledge, reason over it, and answer questions comparably physicians has long been viewed as one such grand challenge. Large language models (LLMs) catalyzed significant progress question answering; Med-PaLM was the first model exceed a "passing" score US Medical Licensing Examination (USMLE) style with of 67.2% on MedQA dataset....
BackgroundMedicine is inherently multimodal, requiring the simultaneous interpretation and integration of insights between many data modalities spanning text, imaging, genomics, more. Generalist biomedical artificial intelligence systems that flexibly encode, integrate, interpret these might better enable impactful applications ranging from scientific discovery to care delivery.MethodsTo catalyze development models, we curated MultiMedBench, a new multimodal benchmark. MultiMedBench...
Federated learning and analytics are a distributed approach for collaboratively models (or statistics) from decentralized data, motivated by designed privacy protection. The process can be formulated as solving federated optimization problems, which emphasize communication efficiency, data heterogeneity, compatibility with system requirements, other constraints that not primary considerations in problem settings. This paper provides recommendations guidelines on formulating, designing,...
Abstract Privacy protection is paramount in conducting health research. However, studies often rely on data stored a centralized repository, where analysis done with full access to the sensitive underlying content. Recent advances federated learning enable building complex machine-learned models that are trained distributed fashion. These techniques facilitate calculation of research study endpoints such private never leaves given device or healthcare system. We show—on diverse set single...
Abstract Large language models (LLMs) represent a major advance in artificial intelligence and, particular, toward the goal of human-like general intelligence. It is sometimes claimed, though, that machine learning “just statistics,” hence that, this grander ambition, progress AI illusory. Here I take contrary view LLMs have great deal to teach us about nature language, understanding, intelligence, sociality, and personhood. Specifically: statistics do amount any falsifiable sense....
Large language models (LLMs) have shown promise in medical question answering, with Med-PaLM being the first to exceed a 'passing' score United States Medical Licensing Examination style questions. However, challenges remain long-form answering and handling real-world workflows. Here, we present 2, which bridges these gaps combination of base LLM improvements, domain fine-tuning new strategies for improving reasoning grounding through ensemble refinement chain retrieval. 2 scores up 86.5% on...
This paper focuses on the opportunities and ethical societal risks posed by advanced AI assistants. We define assistants as artificial agents with natural language interfaces, whose function is to plan execute sequences of actions behalf a user, across one or more domains, in line user's expectations. The starts considering technology itself, providing an overview assistants, their technical foundations potential range applications. It then explores questions around value alignment,...
A spiking neuron "computes" by transforming a complex dynamical input into train of action potentials, or spikes. The computation performed the can be formulated as dimensional reduction, feature detection, followed nonlinear decision function over low-dimensional space. Generalizations reverse correlation technique with white noise provide numerical strategy for extracting relevant features from experimental data, and information theory used to evaluate quality approximation. We apply these...
Large language models (LLMs) have demonstrated impressive capabilities in natural understanding and generation, but the quality bar for medical clinical applications is high. Today, attempts to assess models' knowledge typically rely on automated evaluations limited benchmarks. There no standard evaluate model predictions reasoning across a breadth of tasks. To address this, we present MultiMedQA, benchmark combining six existing open question answering datasets spanning professional exams,...
The computation performed by a neuron can be formulated as combination of dimensional reduction in stimulus space and the nonlinearity inherent spiking output. White noise reverse correlation (the spike-triggered average covariance) are often used experimental neuroscience to "ask" neurons which dimensions they sensitive characterize response. In this article, we apply simplest model with temporal dynamics-the leaky integrate-and-fire model-and find that for even simple case, standard...
To improve real-world applications of machine learning, experienced modelers develop intuition about their datasets, models, and how the two interact. Manual inspection raw data - representative samples, outliers, misclassifications is an essential tool in a) identifying fixing problems data, b) generating new modeling hypotheses, c) assigning or refining human-provided labels. However, manual problematic for privacy sensitive such as those representing behavior individuals. Furthermore,...
Medicine is inherently multimodal, with rich data modalities spanning text, imaging, genomics, and more. Generalist biomedical artificial intelligence (AI) systems that flexibly encode, integrate, interpret this at scale can potentially enable impactful applications ranging from scientific discovery to care delivery. To the development of these models, we first curate MultiMedBench, a new multimodal benchmark. MultiMedBench encompasses 14 diverse tasks such as medical question answering,...
This paper examines the extent to which large language models (LLMs) have developed higher-order theory of mind (ToM); human ability reason about multiple mental and emotional states in a recursive manner (e.g. I think that you believe she knows). builds on prior work by introducing handwritten test suite -- Multi-Order Theory Mind Q&A using it compare performance five LLMs newly gathered adult benchmark. We find GPT-4 Flan-PaLM reach adult-level near ToM tasks overall, exceeds 6th order...
White noise methods are a powerful tool for characterizing the computation performed by neural systems. These allow one to identify feature or features that system extracts from complex input and determine how these combined drive system's spiking response. have also been applied characterize input-output relations of single neurons driven synaptic inputs, simulated direct current injection. To interpret results white analysis neurons, we would like understand obtained space neuron maps onto...
In this wide‐ranging essay, the leader of Google’s Seattle AI group and founder Artists Machine Intelligence program discusses long‐standing complex relationship between art technology. The transformation artistic practice theory that attended 19th century photographic revolution is explored as a parallel for current in machine intelligence, which promises not only to mechanize (or democratize) means reproduction, but also production.
In this talk, Blaise Agüera y Arcas synthesizes the key concepts of his Antikythera book, What is Intelligence?. He argues that intelligence fundamentally prediction—probability future given past—a function applicable to both life and AI. This drive fuels evolution complexity. Modern AI represents genuine intelligence, marking a major evolutionary transition.