D. Sculley's research while affiliated with Google Inc. and other places

The generative AI revolution in recent years has been spurred by an expansion in compute power and data quantity, which together enable extensive pre-training of powerful text-to-image (T2I) models. With their greater capabilities to generate realistic and creative content, these T2I models like DALL-E, MidJourney, Imagen or Stable Diffusion are reaching ever wider audiences. Any unsafe behaviors inherited from pretraining on uncurated internet-scraped datasets thus have the potential to cause wide-reaching harm, for example, through generated images which are violent, sexually explicit, or contain biased and derogatory stereotypes. Despite this risk of harm, we lack systematic and structured evaluation datasets to scrutinize model behavior, especially adversarial attacks that bypass existing safety filters. A typical bottleneck in safety evaluation is achieving a wide coverage of different types of challenging examples in the evaluation set, i.e., identifying 'unknown unknowns' or long-tail problems. To address this need, we introduce the Adversarial Nibbler challenge. The goal of this challenge is to crowdsource a diverse set of failure modes and reward challenge participants for successfully finding safety vulnerabilities in current state-of-the-art T2I models. Ultimately, we aim to provide greater awareness of these issues and assist developers in improving the future safety and reliability of generative AI models. Adversarial Nibbler is a data-centric challenge, part of the DataPerf challenge suite, organized and supported by Kaggle and MLCommons.

A recent trend in artificial intelligence is the use of pretrained models for language and vision tasks, which have achieved extraordinary performance but also puzzling failures. Probing these models' abilities in diverse ways is therefore critical to the field. In this paper, we explore the reliability of models, where we define a reliable model as one that not only achieves strong predictive performance but also performs well consistently over many decision-making tasks involving uncertainty (e.g., selective prediction, open set recognition), robust generalization (e.g., accuracy and proper scoring rules such as log-likelihood on in- and out-of-distribution datasets), and adaptation (e.g., active learning, few-shot uncertainty). We devise 10 types of tasks over 40 datasets in order to evaluate different aspects of reliability on both vision and language domains. To improve reliability, we developed ViT-Plex and T5-Plex, pretrained large model extensions for vision and language modalities, respectively. Plex greatly improves the state-of-the-art across reliability tasks, and simplifies the traditional protocol as it improves the out-of-the-box performance and does not require designing scores or tuning the model for each task. We demonstrate scaling effects over model sizes up to 1B parameters and pretraining dataset sizes up to 4B examples. We also demonstrate Plex's capabilities on challenging tasks including zero-shot open set recognition, active learning, and uncertainty in conversational language understanding.

Understanding the relationship between amino acid sequence and protein function is a long-standing challenge with far-reaching scientific and translational implications. State-of-the-art alignment-based techniques cannot predict function for one-third of microbial protein sequences, hampering our ability to exploit data from diverse organisms. Here, we train deep learning models to accurately predict functional annotations for unaligned amino acid sequences across rigorous benchmark assessments built from the 17,929 families of the protein families database Pfam. The models infer known patterns of evolutionary substitutions and learn representations that accurately cluster sequences from unseen families. Combining deep models with existing methods significantly improves remote homology detection, suggesting that the deep models learn complementary information. This approach extends the coverage of Pfam by >9.5%, exceeding additions made over the last decade, and predicts function for 360 human reference proteome proteins with no previous Pfam annotation. These results suggest that deep learning models will be a core component of future protein annotation tools. A deep learning model predicts protein functional annotations for unaligned amino acid sequences.

Machine learning (ML) systems often exhibit unexpectedly poor behavior when they are deployed in real-world domains. We identify underspecification in ML pipelines as a key reason for these failures. An ML pipeline is the full procedure followed to train and validate a predictor. Such a pipeline is underspecified when it can return many distinct predictors with equivalently strong test performance. Underspecification is common in modern ML pipelines that primarily validate predictors on held-out data that follow the same distribution as the training data. Predictors returned by underspecified pipelines are often treated as equivalent based on their training domain performance, but we show here that such predictors can behave very differently in deployment domains. This ambiguity can lead to instability and poor model behavior in practice, and is a distinct failure mode from previously identified issues arising from structural mismatch between training and deployment domains. We provide evidence that underspecfication has substantive implications for practical ML pipelines, using examples from computer vision, medical imaging, natural language processing, clinical risk prediction based on electronic health records, and medical genomics. Our results show the need to explicitly account for underspecification in modeling pipelines that are intended for real-world deployment in any domain.

High-quality estimates of uncertainty and robustness are crucial for numerous real-world applications, especially for deep learning which underlies many deployed ML systems. The ability to compare techniques for improving these estimates is therefore very important for research and practice alike. Yet, competitive comparisons of methods are often lacking due to a range of reasons, including: compute availability for extensive tuning, incorporation of sufficiently many baselines, and concrete documentation for reproducibility. In this paper we introduce Uncertainty Baselines: high-quality implementations of standard and state-of-the-art deep learning methods on a variety of tasks. As of this writing, the collection spans 19 methods across 9 tasks, each with at least 5 metrics. Each baseline is a self-contained experiment pipeline with easily reusable and extendable components. Our goal is to provide immediate starting points for experimentation with new methods or applications. Additionally we provide model checkpoints, experiment outputs as Python notebooks, and leaderboards for comparing results. Code available at https://github.com/google/uncertainty-baselines.

Genome-wide association studies (GWASs) require accurate cohort phenotyping, but expert labeling can be costly, time intensive, and variable. Here, we develop a machine learning (ML) model to predict glaucomatous optic nerve head features from color fundus photographs. We used the model to predict vertical cup-to-disc ratio (VCDR), a diagnostic parameter and cardinal endophenotype for glaucoma, in 65,680 Europeans in the UK Biobank (UKB). A GWAS of ML-based VCDR identified 299 independent genome-wide significant (GWS; p ≤ 5 × 10⁻⁸) hits in 156 loci. The ML-based GWAS replicated 62 of 65 GWS loci from a recent VCDR GWAS in the UKB for which two ophthalmologists manually labeled images for 67,040 Europeans. The ML-based GWAS also identified 93 novel loci, significantly expanding our understanding of the genetic etiologies of glaucoma and VCDR. Pathway analyses support the biological significance of the novel hits to VCDR: select loci near genes involved in neuronal and synaptic biology or harboring variants are known to cause severe Mendelian ophthalmic disease. Finally, the ML-based GWAS results significantly improve polygenic prediction of VCDR and primary open-angle glaucoma in the independent EPIC-Norfolk cohort.

Genome-wide association studies (GWAS) require accurate cohort phenotyping, but expert labeling can be costly, time-intensive, and variable. Here we develop a machine learning (ML) model to predict glaucomatous optic nerve head features from color fundus photographs. We used the model to predict vertical cup-to-disc ratio (VCDR), a diagnostic parameter and cardinal endophenotype for glaucoma, in 65,680 Europeans in the UK Biobank (UKB). A GWAS of ML-based VCDR identified 299 independent genome-wide significant (GWS; $P\leq5\times10^{-8}$) hits in 156 loci. The ML-based GWAS replicated 62 of 65 GWS loci from a recent VCDR GWAS in the UKB for which two ophthalmologists manually labeled images for 67,040 Europeans. The ML-based GWAS also identified 92 novel loci, significantly expanding our understanding of the genetic etiologies of glaucoma and VCDR. Pathway analyses support the biological significance of the novel hits to VCDR, with select loci near genes involved in neuronal and synaptic biology or known to cause severe Mendelian ophthalmic disease. Finally, the ML-based GWAS results significantly improve polygenic prediction of VCDR and primary open-angle glaucoma in the independent EPIC-Norfolk cohort.

ML models often exhibit unexpectedly poor behavior when they are deployed in real-world domains. We identify underspecification as a key reason for these failures. An ML pipeline is underspecified when it can return many predictors with equivalently strong held-out performance in the training domain. Underspecification is common in modern ML pipelines, such as those based on deep learning. Predictors returned by underspecified pipelines are often treated as equivalent based on their training domain performance, but we show here that such predictors can behave very differently in deployment domains. This ambiguity can lead to instability and poor model behavior in practice, and is a distinct failure mode from previously identified issues arising from structural mismatch between training and deployment domains. We show that this problem appears in a wide variety of practical ML pipelines, using examples from computer vision, medical imaging, natural language processing, clinical risk prediction based on electronic health records, and medical genomics. Our results show the need to explicitly account for underspecification in modeling pipelines that are intended for real-world deployment in any domain.

Covariate shift has been shown to sharply degrade both predictive accuracy and the calibration of uncertainty estimates for deep learning models. This is worrying, because covariate shift is prevalent in a wide range of real world deployment settings. However, in this paper, we note that frequently there exists the potential to access small unlabeled batches of the shifted data just before prediction time. This interesting observation enables a simple but surprisingly effective method which we call prediction-time batch normalization, which significantly improves model accuracy and calibration under covariate shift. Using this one line code change, we achieve state-of-the-art on recent covariate shift benchmarks and an mCE of 60.28\% on the challenging ImageNet-C dataset; to our knowledge, this is the best result for any model that does not incorporate additional data augmentation or modification of the training pipeline. We show that prediction-time batch normalization provides complementary benefits to existing state-of-the-art approaches for improving robustness (e.g. deep ensembles) and combining the two further improves performance. Our findings are supported by detailed measurements of the effect of this strategy on model behavior across rigorous ablations on various dataset modalities. However, the method has mixed results when used alongside pre-training, and does not seem to perform as well under more natural types of dataset shift, and is therefore worthy of additional study. We include links to the data in our figures to improve reproducibility, including a Python notebooks that can be run to easily modify our analysis at https://colab.research.google.com/drive/11N0wDZnMQQuLrRwRoumDCrhSaIhkqjof.