Foundation Models (FMs) trained on Electronic Health Records (EHRs) have achieved state-of-the-art results on numerous clinical prediction tasks. However, prior EHR FMs typically have context windows of $<$1k tokens, which prevents them from modeling full patient EHRs which can exceed 10k's of events. For making clinical predictions, both model performance and robustness to the unique properties of EHR data are crucial. Recent advancements in subquadratic long-context architectures (e.g. Mamba) offer a promising solution. However, their application to EHR data has not been well-studied. We address this gap by presenting the first systematic evaluation of the effect of context length on modeling EHR data. We find that longer context models improve predictive performance -- our Mamba-based model surpasses the prior state-of-the-art on 9/14 tasks on the EHRSHOT prediction benchmark. Additionally, we measure robustness to three unique, previously underexplored properties of EHR data: (1) the prevalence of ``copy-forwarded" diagnoses which create artificial token repetition in EHR sequences; (2) the irregular time intervals between EHR events which can lead to a wide range of timespans within a context window; and (3) the natural increase in disease complexity over time which makes later tokens in the EHR harder to predict than earlier ones. Stratifying our EHRSHOT results, we find that higher levels of each property correlate negatively with model performance (e.g., a 14% higher Brier loss between the least and most irregular patients), but that longer context models are more robust to more extreme levels of these properties. Our work highlights the potential for using long-context architectures to model EHR data, and offers a case study on how to identify and quantify new challenges in modeling sequential data motivated by domains outside of natural language. We release all of our model checkpoints and code.

Health AI suffers from a systemic reproducibility crisis that irreparably hinders research across both academia and industry [4,5].One key tool poised to solve this crisis is the Medical Event Data Standard (MEDS), a comprehensive data format and open-source ecosystem designed to enhance reproducibility and interoperability of AI research using longitudinal Electronic Health Records (EHR) [6].Currently adopted by over 15 institutions globally, MEDS encompasses various open-source tools, published models, and data processing pipelines, enabling streamlined model development and robust benchmarking.In this tutorial, participants will gain key hands-on experience in working with the MEDS format to perform efficient, reproducible, state-of-the-art AI research over real health data.Participants will transform data into the MEDS format, preprocess data, build predictive models, and contribute to the decentralized MEDS-DEV benchmarking platform.Interactive exercises using Jupyter notebooks will provide hands-on experience and practical skills for reproducible health AI research.Attendees will leave equipped

LLM-based agent applications have shown increasingly remarkable capabilities in complex workflows but incur substantial costs and latency due to extensive planning and reasoning requirements. Existing LLM caching techniques (like context caching and semantic caching), primarily designed for serving chatbots, are insufficient for agent applications where outputs depend on external data and environmental contexts. We propose **Agentic Plan Caching (APC)**, a novel **test-time memory** that extracts, stores, adapts, and reuses structured plan templates from planning stages of agent applications across semantically similar tasks to reduce the cost and latency of serving. Unlike traditional semantic caching, our system extracts plan templates from completed agent executions at test-time, employs keyword extraction to match new requests against cached plans, and utilizes lightweight models to adapt these templates to task-specific plans with contexts. Evaluation across multiple real-world agent applications shows that our system can reduce costs by 50.31\% and latency by 27.28\% on average while maintaining performance, offering a more efficient solution for serving LLM-based agents that complements existing LLM serving infrastructures.

Existing ML benchmarks lack the depth and diversity of annotations needed for evaluating models on business process management (BPM) tasks. BPM is the practice of documenting, measuring, improving, and automating enterprise workflows. However, research has focused almost exclusively on one task -- full end-to-end automation using agents based on multimodal foundation models (FMs) like GPT-4. This focus on automation ignores the reality of how most BPM tools are applied today -- simply documenting the relevant workflow takes 60% of the time of the typical process optimization project. To address this gap we present WONDERBREAD, the first benchmark for evaluating multimodal FMs on BPM tasks beyond automation. Our contributions are: (1) a dataset containing 2928 documented workflow demonstrations; (2) 6 novel BPM tasks sourced from real-world applications ranging from workflow documentation to knowledge transfer to process improvement; and (3) an automated evaluation harness. Our benchmark shows that while state-of-the-art FMs can automatically generate documentation (e.g. recalling 88% of the steps taken in a video demonstration of a workflow), they struggle to re-apply that knowledge towards finer-grained validation of workflow completion (F1 < 0.3). We hope WONDERBREAD encourages the development of more "human-centered" AI tooling for enterprise applications and furthers the exploration of multimodal FMs for the broader universe of BPM tasks. We publish our dataset and experiments here: https://github.com/HazyResearch/wonderbread

Automating enterprise workflows could unlock $4 trillion/year in productivity gains. Despite being of interest to the data management community for decades, the ultimate vision of end-to-end workflow automation has remained elusive. Current solutions rely on process mining and robotic process automation (RPA), in which a bot is hard-coded to follow a set of predefined rules for completing a workflow. Through case studies of a hospital and large B2B enterprise, we find that the adoption of RPA has been inhibited by high set-up costs (12--18 months), unreliable execution (60% initial accuracy), and burdensome maintenance (requiring multiple FTEs). Multimodal foundation models (FMs) such as GPT-4 offer a promising new approach for end-to-end workflow automation given their generalized reasoning and planning abilities. To study these capabilities we propose ECLAIR, a system to automate enterprise workflows with minimal human supervision. We conduct initial experiments showing that multimodal FMs can address the limitations of traditional RPA with (1) near-human-level understanding of workflows (93% accuracy on a workflow understanding task) and (2) instant set-up with minimal technical barrier (based solely on a natural language description of a workflow, ECLAIR achieves end-to-end completion rates of 40%). We identify human-AI collaboration, validation, and self-improvement as open challenges, and suggest ways they can be solved with data management techniques.

Genomic (DNA) sequences encode an enormous amount of information for gene regulation and protein synthesis. Similar to natural language models, researchers have proposed foundation models in genomics to learn generalizable features from unlabeled genome data that can then be fine-tuned for downstream tasks such as identifying regulatory elements. Due to the quadratic scaling of attention, previous Transformer-based genomic models have used 512 to 4k tokens as context (<0.001% of the human genome), significantly limiting the modeling of long-range interactions in DNA. In addition, these methods rely on tokenizers or fixed k-mers to aggregate meaningful DNA units, losing single nucleotide resolution (i.e. DNA "characters") where subtle genetic variations can completely alter protein function via single nucleotide polymorphisms (SNPs). Recently, Hyena, a large language model based on implicit convolutions was shown to match attention in quality while allowing longer context lengths and lower time complexity. Leveraging Hyena’s new long-range capabilities, we present HyenaDNA, a genomic foundation model pretrained on the human reference genome with context lengths of up to 1 million tokens at the single nucleotide-level – an up to 500x increase over previous dense attention-based models. HyenaDNA scales sub-quadratically in sequence length (training up to 160x faster than Transformer), uses single nucleotide tokens, and has full global context at each layer. We explore what longer context enables - including the first use of in-context learning in genomics for simple adaptation to novel tasks without updating pretrained model weights. On fine-tuned benchmarks from the Nucleotide Transformer, HyenaDNA reaches state-of-the-art (SotA) on 12 of 18 datasets using a model with orders of magnitude less parameters and pretraining data.1 On the GenomicBenchmarks, HyenaDNA surpasses SotA on 7 of 8 datasets on average by +10 accuracy points. Code at https://github.com/HazyResearch/hyena-dna.

While the general machine learning (ML) community has benefited from public datasets, tasks, and models, the progress of ML in healthcare has been hampered by a lack of such shared assets. The success of foundation models creates new challenges for healthcare ML by requiring access to shared pretrained models to validate performance benefits. We help address these challenges through three contributions. First, we publish a new dataset, EHRSHOT, which contains de-identified structured data from the electronic health records (EHRs) of 6,739 patients from Stanford Medicine. Unlike MIMIC-III/IV and other popular EHR datasets, EHRSHOT is longitudinal and not restricted to ICU/ED patients. Second, we publish the weights of CLMBR-T-base, a 141M parameter clinical foundation model pretrained on the structured EHR data of 2.57M patients. We are one of the first to fully release such a model for coded EHR data; in contrast, most prior models released for clinical data (e.g. GatorTron, ClinicalBERT) only work with unstructured text and cannot process the rich, structured data within an EHR. We provide an end-to-end pipeline for the community to validate and build upon its performance. Third, we define 15 few-shot clinical prediction tasks, enabling evaluation of foundation models on benefits such as sample efficiency and task adaptation. Our model and dataset are available via a research data use agreement from here: https://stanfordaimi.azurewebsites.net/. Code to reproduce our results is available here: https://github.com/som-shahlab/ehrshot-benchmark.

Despite the creation of thousands of machine learning (ML) models, the promise of improving patient care with ML remains largely unrealized. Adoption into clinical practice is lagging, in large part due to disconnects between how ML practitioners evaluate models and what is required for their successful integration into care delivery. Models are just one component of care delivery workflows whose constraints determine clinicians' abilities to act on models' outputs. However, methods to evaluate the usefulness of models in the context of their corresponding workflows are currently limited. To bridge this gap we developed APLUS, a reusable framework for quantitatively assessing via simulation the utility gained from integrating a model into a clinical workflow. We describe the APLUS simulation engine and workflow specification language, and apply it to evaluate a novel ML-based screening pathway for detecting peripheral artery disease at Stanford Health Care.

The pathogenesis of many inflammatory diseases is a coordinated process involving metabolic dysfunctions and immune response-usually modulated by the production of cytokines and associated inflammatory molecules. In this work, we seek to understand how genes involved in pathogenesis which are often not associated with the immune system in an obvious way communicate with the immune system. We have embedded a network of human protein-protein interactions (PPI) from the STRING database with 14,707 human genes using feature learning that captures high confidence edges. We have found that our predicted Association Scores derived from the features extracted from STRING's high confidence edges are useful for predicting novel connections between genes, thus enabling the construction of a full map of predicted associations for all possible pairs between 14,707 human genes. In particular, we analyzed the pattern of associations for 126 cytokines and found that the six patterns of cytokine interaction with human genes are consistent with their functional classifications. To define the disease-specific roles of cytokines we have collected gene sets for 11,944 diseases from DisGeNET. We used these gene sets to predict disease-specific gene associations with cytokines by calculating the normalized average Association Scores between disease-associated gene sets and the 126 cytokines; this creates a unique profile of inflammatory genes (both known and predicted) for each disease. We validated our predicted cytokine associations by comparing them to known associations for 171 diseases. The predicted cytokine profiles correlate (p-value<0.0003) with the known ones in 95 diseases. We further characterized the profiles of each disease by calculating an "Inflammation Score" that summarizes different modes of immune responses. Finally, by analyzing subnetworks formed between disease-specific pathogenesis genes, hormones, receptors, and cytokines, we identified the key genes responsible for interactions between pathogenesis and inflammatory responses. These genes and the corresponding cytokines used by different immune disorders suggest unique targets for drug discovery.