David Madras

Gemini Team, Petko Georgiev, Ving Ian Lei et al. · deepmind, mila

In this report, we introduce the Gemini 1.5 family of models, representing the next generation of highly compute-efficient multimodal models capable of recalling and reasoning over fine-grained information from millions of tokens of context, including multiple long documents and hours of video and audio. The family includes two new models: (1) an updated Gemini 1.5 Pro, which exceeds the February version on the great majority of capabilities and benchmarks; (2) Gemini 1.5 Flash, a more lightweight variant designed for efficiency with minimal regression in quality. Gemini 1.5 models achieve near-perfect recall on long-context retrieval tasks across modalities, improve the state-of-the-art in long-document QA, long-video QA and long-context ASR, and match or surpass Gemini 1.0 Ultra's state-of-the-art performance across a broad set of benchmarks. Studying the limits of Gemini 1.5's long-context ability, we find continued improvement in next-token prediction and near-perfect retrieval (>99%) up to at least 10M tokens, a generational leap over existing models such as Claude 3.0 (200k) and GPT-4 Turbo (128k). Finally, we highlight real-world use cases, such as Gemini 1.5 collaborating with professionals on completing their tasks achieving 26 to 75% time savings across 10 different job categories, as well as surprising new capabilities of large language models at the frontier; when given a grammar manual for Kalamang, a language with fewer than 200 speakers worldwide, the model learns to translate English to Kalamang at a similar level to a person who learned from the same content.

Minghe Shen, Ananth Balashankar, Adam Fisch et al.

The ability to rigorously estimate the failure rates of large language models (LLMs) is a prerequisite for their safe deployment. Currently, however, practitioners often face a tradeoff between expensive human gold standards and potentially severely-biased automatic annotation schemes such as "LLM-as-a-Judge" labeling. In this paper, we propose a new, practical, and efficient approach to LLM failure rate estimation based on constrained maximum-likelihood estimation (MLE). Our method integrates three distinct signal sources: (i) a small, high-quality human-labeled calibration set, (ii) a large corpus of LLM-judge annotations, and, most importantly, (iii) additional side information via domain-specific constraints derived from known bounds on judge performance statistics. We validate our approach through a comprehensive empirical study, benchmarking it against state-of-the-art baselines like Prediction-Powered Inference (PPI). Across diverse experimental regimes -- spanning varying judge accuracies, calibration set sizes, and LLM failure rates -- our constrained MLE consistently delivers more accurate and lower-variance estimates than existing methods. By moving beyond the "black-box" use of automated judges to a flexible framework, we provide a principled, interpretable, and scalable pathway towards LLM failure-rate certification.

Gheorghe Comanici, Eric Bieber, Mike Schaekermann et al. · amazon-science, baidu

In this report, we introduce the Gemini 2.X model family: Gemini 2.5 Pro and Gemini 2.5 Flash, as well as our earlier Gemini 2.0 Flash and Flash-Lite models. Gemini 2.5 Pro is our most capable model yet, achieving SoTA performance on frontier coding and reasoning benchmarks. In addition to its incredible coding and reasoning skills, Gemini 2.5 Pro is a thinking model that excels at multimodal understanding and it is now able to process up to 3 hours of video content. Its unique combination of long context, multimodal and reasoning capabilities can be combined to unlock new agentic workflows. Gemini 2.5 Flash provides excellent reasoning abilities at a fraction of the compute and latency requirements and Gemini 2.0 Flash and Flash-Lite provide high performance at low latency and cost. Taken together, the Gemini 2.X model generation spans the full Pareto frontier of model capability vs cost, allowing users to explore the boundaries of what is possible with complex agentic problem solving.

Jiachen Zhao, Zhun Deng, David Madras et al.

As the number of large language models (LLMs) released to the public grows, there is a pressing need to understand the safety implications associated with these models learning from third-party custom finetuning data. We explore the behavior of LLMs finetuned on noisy custom data containing unsafe content, represented by datasets that contain biases, toxicity, and harmfulness, finding that while aligned LLMs can readily learn this unsafe content, they also tend to forget it more significantly than other examples when subsequently finetuned on safer content. Drawing inspiration from the discrepancies in forgetting, we introduce the "ForgetFilter" algorithm, which filters unsafe data based on how strong the model's forgetting signal is for that data. We demonstrate that the ForgetFilter algorithm ensures safety in customized finetuning without compromising downstream task performance, unlike sequential safety finetuning. ForgetFilter outperforms alternative strategies like replay and moral self-correction in curbing LLMs' ability to assimilate unsafe content during custom finetuning, e.g. 75% lower than not applying any safety measures and 62% lower than using self-correction in toxicity score.

Benjamin Eyre, David Madras

The availability of machine learning systems that can effectively perform arbitrary tasks has led to synthetic labels from these systems being used in applications of statistical inference, such as data analysis or model evaluation. The Prediction Powered Inference (PPI) framework provides a way of leveraging both a large pool of pseudo-labelled data and a small sample with real, high-quality labels to produce a low-variance, unbiased estimate of the quantity being evaluated for. Most work on PPI considers a relatively sizable set of labelled samples, which can be resource intensive to obtain. However, we find that when labelled data is scarce, the PPI++ method can perform even worse than classical inference. We analyze this phenomenon by relating PPI++ to ordinary least squares regression, which also experiences high variance with small sample sizes, and use this regression framework to better understand the efficacy of PPI. Motivated by this, we present two new PPI-based techniques that leverage robust regressors to produce even lower variance estimators in the few-label regime.

Benjamin Eyre, Elliot Creager, David Madras et al. · deepmind

Designing deep neural network classifiers that perform robustly on distributions differing from the available training data is an active area of machine learning research. However, out-of-distribution generalization for regression-the analogous problem for modeling continuous targets-remains relatively unexplored. To tackle this problem, we return to first principles and analyze how the closed-form solution for Ordinary Least Squares (OLS) regression is sensitive to covariate shift. We characterize the out-of-distribution risk of the OLS model in terms of the eigenspectrum decomposition of the source and target data. We then use this insight to propose a method for adapting the weights of the last layer of a pre-trained neural regression model to perform better on input data originating from a different distribution. We demonstrate how this lightweight spectral adaptation procedure can improve out-of-distribution performance for synthetic and real-world datasets.

Stephen R. Pfohl, Natalie Harris, Chirag Nagpal et al.

Disaggregated evaluation across subgroups is critical for assessing the fairness of machine learning models, but its uncritical use can mislead practitioners. We show that equal performance across subgroups is an unreliable measure of fairness when data are representative of the relevant populations but reflective of real-world disparities. Furthermore, when data are not representative due to selection bias, both disaggregated evaluation and alternative approaches based on conditional independence testing may be invalid without explicit assumptions regarding the bias mechanism. We use causal graphical models to characterize fairness properties and metric stability across subgroups under different data generating processes. Our framework suggests complementing disaggregated evaluations with explicit causal assumptions and analysis to control for confounding and distribution shift, including conditional independence testing and weighted performance estimation. These findings have broad implications for how practitioners design and interpret model assessments given the ubiquity of disaggregated evaluation.

David Madras, Joshua Safyan, Qiuyi et al.

Many prompt engineering techniques have been successful in practice, even when optimizing over a large prompt space with with a small amount of task-specific data. Recent work has partially explained this success by showing generalization bounds which apply PAC-Bayes theory to the discrete prompt space, but they are non-vacuous only in data-rich scenarios. We argue that such widespread success can be more fully explained through more carefully considering data- or distribution-dependent perplexity, which acts as an effective prior and steers the optimization towards prompts that are more ``natural'' for the task at hand. We derive novel generalization bounds that are non-vacuous for data-scarce prompt optimization via more useful priors, formally analyzing how perplexity regularization tightens these bounds by limiting exploration. Empirically, we explore both the bounds' effectiveness and the practical benefits of perplexity regularization in improving prompt generalization.

Zhun Deng, Thomas P Zollo, Benjamin Eyre et al.

As machine learning models grow increasingly competent, their predictions can supplement scarce or expensive data in various important domains. In support of this paradigm, algorithms have emerged to combine a small amount of high-fidelity observed data with a much larger set of imputed model outputs to estimate some quantity of interest. Yet current hybrid-inference tools target only means or single quantiles, limiting their applicability for many critical domains and use cases. We present QuEst, a principled framework to merge observed and imputed data to deliver point estimates and rigorous confidence intervals for a wide family of quantile-based distributional measures. QuEst covers a range of measures, from tail risk (CVaR) to population segments such as quartiles, that are central to fields such as economics, sociology, education, medicine, and more. We extend QuEst to multidimensional metrics, and introduce an additional optimization technique to further reduce variance in this and other hybrid estimators. We demonstrate the utility of our framework through experiments in economic modeling, opinion polling, and language model auto-evaluation.

Hansa Srinivasan, Candice Schumann, Aradhana Sinha et al.

Capturing the diversity of people in images is challenging: recent literature tends to focus on diversifying one or two attributes, requiring expensive attribute labels or building classifiers. We introduce a diverse people image ranking method which more flexibly aligns with human notions of people diversity in a less prescriptive, label-free manner. The Perception-Aligned Text-derived Human representation Space (PATHS) aims to capture all or many relevant features of people-related diversity, and, when used as the representation space in the standard Maximal Marginal Relevance (MMR) ranking algorithm, is better able to surface a range of types of people-related diversity (e.g. disability, cultural attire). PATHS is created in two stages. First, a text-guided approach is used to extract a person-diversity representation from a pre-trained image-text model. Then this representation is fine-tuned on perception judgments from human annotators so that it captures the aspects of people-related similarity that humans find most salient. Empirical results show that the PATHS method achieves diversity better than baseline methods, according to side-by-side ratings from human annotators.

Gemini Team, Rohan Anil, Sebastian Borgeaud et al.

This report introduces a new family of multimodal models, Gemini, that exhibit remarkable capabilities across image, audio, video, and text understanding. The Gemini family consists of Ultra, Pro, and Nano sizes, suitable for applications ranging from complex reasoning tasks to on-device memory-constrained use-cases. Evaluation on a broad range of benchmarks shows that our most-capable Gemini Ultra model advances the state of the art in 30 of 32 of these benchmarks - notably being the first model to achieve human-expert performance on the well-studied exam benchmark MMLU, and improving the state of the art in every one of the 20 multimodal benchmarks we examined. We believe that the new capabilities of the Gemini family in cross-modal reasoning and language understanding will enable a wide variety of use cases. We discuss our approach toward post-training and deploying Gemini models responsibly to users through services including Gemini, Gemini Advanced, Google AI Studio, and Cloud Vertex AI.

David Madras, Richard Zemel

Deep learning systems frequently fail at out-of-context (OOC) prediction, the problem of making reliable predictions on uncommon or unusual inputs or subgroups of the training distribution. To this end, a number of benchmarks for measuring OOC performance have recently been introduced. In this work, we introduce a framework unifying the literature on OOC performance measurement, and demonstrate how rich auxiliary information can be leveraged to identify candidate sets of OOC examples in existing datasets. We present NOOCh: a suite of naturally-occurring "challenge sets", and show how varying notions of context can be used to probe specific OOC failure modes. Experimentally, we explore the tradeoffs between various learning approaches on these challenge sets and demonstrate how the choices made in designing OOC benchmarks can yield varying conclusions.

Robert Adragna, Elliot Creager, David Madras et al.

Robustness is of central importance in machine learning and has given rise to the fields of domain generalization and invariant learning, which are concerned with improving performance on a test distribution distinct from but related to the training distribution. In light of recent work suggesting an intimate connection between fairness and robustness, we investigate whether algorithms from robust ML can be used to improve the fairness of classifiers that are trained on biased data and tested on unbiased data. We apply Invariant Risk Minimization (IRM), a domain generalization algorithm that employs a causal discovery inspired method to find robust predictors, to the task of fairly predicting the toxicity of internet comments. We show that IRM achieves better out-of-distribution accuracy and fairness than Empirical Risk Minimization (ERM) methods, and analyze both the difficulties that arise when applying IRM in practice and the conditions under which IRM will likely be effective in this scenario. We hope that this work will inspire further studies of how robust machine learning methods relate to algorithmic fairness.

Sindy Löwe, David Madras, Richard Zemel et al.

On time-series data, most causal discovery methods fit a new model whenever they encounter samples from a new underlying causal graph. However, these samples often share relevant information which is lost when following this approach. Specifically, different samples may share the dynamics which describe the effects of their causal relations. We propose Amortized Causal Discovery, a novel framework that leverages such shared dynamics to learn to infer causal relations from time-series data. This enables us to train a single, amortized model that infers causal relations across samples with different underlying causal graphs, and thus leverages the shared dynamics information. We demonstrate experimentally that this approach, implemented as a variational model, leads to significant improvements in causal discovery performance, and show how it can be extended to perform well under added noise and hidden confounding.

David Madras, James Atwood, Alex D'Amour

We present local ensembles, a method for detecting underspecification -- when many possible predictors are consistent with the training data and model class -- at test time in a pre-trained model. Our method uses local second-order information to approximate the variance of predictions across an ensemble of models from the same class. We compute this approximation by estimating the norm of the component of a test point's gradient that aligns with the low-curvature directions of the Hessian, and provide a tractable method for estimating this quantity. Experimentally, we show that our method is capable of detecting when a pre-trained model is underspecified on test data, with applications to out-of-distribution detection, detecting spurious correlates, and active learning.

Elliot Creager, David Madras, Toniann Pitassi et al.

In many application areas---lending, education, and online recommenders, for example---fairness and equity concerns emerge when a machine learning system interacts with a dynamically changing environment to produce both immediate and long-term effects for individuals and demographic groups. We discuss causal directed acyclic graphs (DAGs) as a unifying framework for the recent literature on fairness in such dynamical systems. We show that this formulation affords several new directions of inquiry to the modeler, where causal assumptions can be expressed and manipulated. We emphasize the importance of computing interventional quantities in the dynamical fairness setting, and show how causal assumptions enable simulation (when environment dynamics are known) and off-policy estimation (when dynamics are unknown) of intervention on short- and long-term outcomes, at both the group and individual levels.

Elliot Creager, David Madras, Jörn-Henrik Jacobsen et al.

We consider the problem of learning representations that achieve group and subgroup fairness with respect to multiple sensitive attributes. Taking inspiration from the disentangled representation learning literature, we propose an algorithm for learning compact representations of datasets that are useful for reconstruction and prediction, but are also \emph{flexibly fair}, meaning they can be easily modified at test time to achieve subgroup demographic parity with respect to multiple sensitive attributes and their conjunctions. We show empirically that the resulting encoder---which does not require the sensitive attributes for inference---enables the adaptation of a single representation to a variety of fair classification tasks with new target labels and subgroup definitions.

David Madras, Elliot Creager, Toniann Pitassi et al.

How do we learn from biased data? Historical datasets often reflect historical prejudices; sensitive or protected attributes may affect the observed treatments and outcomes. Classification algorithms tasked with predicting outcomes accurately from these datasets tend to replicate these biases. We advocate a causal modeling approach to learning from biased data, exploring the relationship between fair classification and intervention. We propose a causal model in which the sensitive attribute confounds both the treatment and the outcome. Building on prior work in deep learning and generative modeling, we describe how to learn the parameters of this causal model from observational data alone, even in the presence of unobserved confounders. We show experimentally that fairness-aware causal modeling provides better estimates of the causal effects between the sensitive attribute, the treatment, and the outcome. We further present evidence that estimating these causal effects can help learn policies that are both more accurate and fair, when presented with a historically biased dataset.

David Madras, Elliot Creager, Toniann Pitassi et al.

In this paper, we advocate for representation learning as the key to mitigating unfair prediction outcomes downstream. Motivated by a scenario where learned representations are used by third parties with unknown objectives, we propose and explore adversarial representation learning as a natural method of ensuring those parties act fairly. We connect group fairness (demographic parity, equalized odds, and equal opportunity) to different adversarial objectives. Through worst-case theoretical guarantees and experimental validation, we show that the choice of this objective is crucial to fair prediction. Furthermore, we present the first in-depth experimental demonstration of fair transfer learning and demonstrate empirically that our learned representations admit fair predictions on new tasks while maintaining utility, an essential goal of fair representation learning.

David Madras, Toniann Pitassi, Richard Zemel

In many machine learning applications, there are multiple decision-makers involved, both automated and human. The interaction between these agents often goes unaddressed in algorithmic development. In this work, we explore a simple version of this interaction with a two-stage framework containing an automated model and an external decision-maker. The model can choose to say "Pass", and pass the decision downstream, as explored in rejection learning. We extend this concept by proposing "learning to defer", which generalizes rejection learning by considering the effect of other agents in the decision-making process. We propose a learning algorithm which accounts for potential biases held by external decision-makers in a system. Experiments demonstrate that learning to defer can make systems not only more accurate but also less biased. Even when working with inconsistent or biased users, we show that deferring models still greatly improve the accuracy and/or fairness of the entire system.

Stephanie Allen, David Madras, Ye Ye et al.

Body-worn video (BWV) cameras are increasingly utilized by police departments to provide a record of police-public interactions. However, large-scale BWV deployment produces terabytes of data per week, necessitating the development of effective computational methods to identify salient changes in video. In work carried out at the 2016 RIPS program at IPAM, UCLA, we present a novel two-stage framework for video change-point detection. First, we employ state-of-the-art machine learning methods including convolutional neural networks and support vector machines for scene classification. We then develop and compare change-point detection algorithms utilizing mean squared-error minimization, forecasting methods, hidden Markov models, and maximum likelihood estimation to identify noteworthy changes. We test our framework on detection of vehicle exits and entrances in a BWV data set provided by the Los Angeles Police Department and achieve over 90% recall and nearly 70% precision -- demonstrating robustness to rapid scene changes, extreme luminance differences, and frequent camera occlusions.