We introduce BSDETECTOR, a method for detecting bad and speculative answers from a pretrained Large Language Model by estimating a numeric confidence score for any output it generated.Our uncertainty quantification technique works for any LLM accessible only via a black-box API, whose training data remains unknown.By expending a bit of extra computation, users of any LLM API can now get the same response as they would ordinarily, as well as a confidence estimate that cautions when not to trust this response.Experiments on both closed and open-form Question-Answer benchmarks reveal that BSDETECTOR more accurately identifies incorrect LLM responses than alternative uncertainty estimation procedures (for both GPT-3 and ChatGPT).By sampling multiple responses from the LLM and considering the one with the highest confidence score, we can additionally obtain more accurate responses from the same LLM, without any extra training steps.In applications involving automated evaluation with LLMs, accounting for our confidence scores leads to more reliable evaluation in both human-in-the-loop and fullyautomated settings (across both GPT 3.5 and 4).... Observed Consistency + Ans...

Real-world deployment of machine learning models is challenging because data evolves over time. While no model can work when data evolves in an arbitrary fashion, if there is some pattern to these changes, we might be able to design methods to address it. This paper addresses situations when data evolves gradually. We introduce a time-varying propensity score that can detect gradual shifts in the distribution of data which allows us to selectively sample past data to update the model---not just similar data from the past like that of a standard propensity score but also data that evolved in a similar fashion in the past. The time-varying propensity score is quite general: we demonstrate different ways of implementing it and evaluate it on a variety of problems ranging from supervised learning (e.g., image classification problems) where data undergoes a sequence of gradual shifts, to reinforcement learning tasks (e.g., robotic manipulation and continuous control) where data shifts as the policy or the task changes.

Although the transformer architecture has come to dominate other models for text and image data, its application to irregularly-spaced longitudinal data has been limited. We introduce a variant of the transformer that enables it to more smoothly impute such functional data. We augment the vanilla transformer with a simple module we call SAND (self-attention on derivatives), which naturally encourages smoothness by modeling the sub-derivative of the imputed curve. On the theoretical front, we prove the number of hidden nodes required by a network with SAND to achieve an $\epsilon$ prediction error bound for functional imputation. Extensive experiments over various types of functional data demonstrate that transformers with SAND produce better imputations than both their standard counterparts as well as transformers augmented with alternative approaches to encode the inductive bias of smoothness. SAND also outperforms standard statistical methods for functional imputation like kernel smoothing and PACE.

Machine learning research has long focused on models rather than datasets, and prominent datasets are used for common ML tasks without regard to the breadth, difficulty, and faithfulness of the underlying problems. Neglecting the fundamental importance of data has given rise to inaccuracy, bias, and fragility in real-world applications, and research is hindered by saturation across existing dataset benchmarks. In response, we present DataPerf, a community-led benchmark suite for evaluating ML datasets and data-centric algorithms. We aim to foster innovation in data-centric AI through competition, comparability, and reproducibility. We enable the ML community to iterate on datasets, instead of just architectures, and we provide an open, online platform with multiple rounds of challenges to support this iterative development. The first iteration of DataPerf contains five benchmarks covering a wide spectrum of data-centric techniques, tasks, and modalities in vision, speech, acquisition, debugging, and diffusion prompting, and we support hosting new contributed benchmarks from the community. The benchmarks, online evaluation platform, and baseline implementations are open source, and the MLCommons Association will maintain DataPerf to ensure long-term benefits to academia and industry.

Quantile regression is a fundamental problem in statistical learning motivated by a need to quantify uncertainty in predictions, or to model a diverse population without being overly reductive. For instance, epidemiological forecasts, cost estimates, and revenue predictions all benefit from being able to quantify the range of possible values accurately. As such, many models have been developed for this problem over many years of research in statistics, machine learning, and related fields. Rather than proposing yet another (new) algorithm for quantile regression we adopt a meta viewpoint: we investigate methods for aggregating any number of conditional quantile models, in order to improve accuracy and robustness. We consider weighted ensembles where weights may vary over not only individual models, but also over quantile levels, and feature values. All of the models we consider in this paper can be fit using modern deep learning toolkits, and hence are widely accessible (from an implementation point of view) and scalable. To improve the accuracy of the predicted quantiles (or equivalently, prediction intervals), we develop tools for ensuring that quantiles remain monotonically ordered, and apply conformal calibration methods. These can be used without any modification of the original library of base models. We also review some basic theory surrounding quantile aggregation and related scoring rules, and contribute a few new results to this literature (for example, the fact that post sorting or post isotonic regression can only improve the weighted interval score). Finally, we provide an extensive suite of empirical comparisons across 34 data sets from two different benchmark repositories.

Emphatic algorithms have shown great promise in stabilizing and improving reinforcement learning by selectively emphasizing the update rule. Although the emphasis fundamentally depends on an interest function which defines the intrinsic importance of each state, most approaches simply adopt a uniform interest over all states (except where a hand-designed interest is possible based on domain knowledge). In this paper, we investigate adaptive methods that allow the interest function to dynamically vary over states and iterations. In particular, we leverage meta-gradients to automatically discover online an interest function that would accelerate the agent’s learning process. Empirical evaluations on a wide range of environments show that adapting the interest is key to provide significant gains. Qualitative analysis indicates that the learned interest function emphasizes states of particular importance, such as bottlenecks, which can be especially useful in a transfer learning setting.

Despite their widespread success, the application of deep neural networks to functional data remains scarce today. The infinite dimensionality of functional data means standard learning algorithms can be applied only after appropriate dimension reduction, typically achieved via basis expansions. Currently, these bases are chosen a priori without the information for the task at hand and thus may not be effective for the designated task. We instead propose to adaptively learn these bases in an end-to-end fashion. We introduce neural networks that employ a new Basis Layer whose hidden units are each basis functions themselves implemented as a micro neural network. Our architecture learns to apply parsimonious dimension reduction to functional inputs that focuses only on information relevant to the target rather than irrelevant variation in the input function. Across numerous classification/regression tasks with functional data, our method empirically outperforms other types of neural networks, and we prove that our approach is statistically consistent with low generalization error.

Image classifiers are typically scored on their test set accuracy, but high accuracy can mask a subtle type of model failure. We find that high scoring convolutional neural networks (CNNs) on popular benchmarks exhibit troubling pathologies that allow them to display high accuracy even in the absence of semantically salient features. When a model provides a high-confidence decision without salient supporting input features, we say the classifier has overinterpreted its input, finding too much class-evidence in patterns that appear nonsensical to humans. Here, we demonstrate that neural networks trained on CIFAR-10 and ImageNet suffer from overinterpretation, and we find models on CIFAR-10 make confident predictions even when 95% of input images are masked and humans cannot discern salient features in the remaining pixel-subsets. We introduce Batched Gradient SIS, a new method for discovering sufficient input subsets for complex datasets, and use this method to show the sufficiency of border pixels in ImageNet for training and testing. Although these patterns portend potential model fragility in real-world deployment, they are in fact valid statistical patterns of the benchmark that alone suffice to attain high test accuracy. Unlike adversarial examples, overinterpretation relies upon unmodified image pixels. We find ensembling and input dropout can each help mitigate overinterpretation.

Reliant on too many experiments to learn good actions, current Reinforcement Learning (RL) algorithms have limited applicability in real-world settings, which can be too expensive to allow exploration. We propose an algorithm for batch RL, where effective policies are learned using only a fixed offline dataset instead of online interactions with the environment. The limited data in batch RL produces inherent uncertainty in value estimates of states/actions that were insufficiently represented in the training data. This leads to particularly severe extrapolation when our candidate policies diverge from one that generated the data. We propose to mitigate this issue via two straightforward penalties: a policy-constraint to reduce this divergence and a value-constraint that discourages overly optimistic estimates. Over a comprehensive set of $32$ continuous-action batch RL benchmarks, our approach compares favorably to state-of-the-art methods, regardless of how the offline data were collected.

Unlike standard prediction tasks, survival analysis requires modeling right censored data, which must be treated with care. While deep neural networks excel in traditional supervised learning, it remains unclear how to best utilize these models in survival analysis. A key question asks which data-generating assumptions of traditional survival models should be retained and which should be made more flexible via the function-approximating capabilities of neural networks. Rather than estimating the survival function targeted by most existing methods, we introduce a Deep Extended Hazard (DeepEH) model to provide a flexible and general framework for deep survival analysis. The extended hazard model includes the conventional Cox proportional hazards and accelerated failure time models as special cases, so DeepEH subsumes the popular Deep Cox proportional hazard (DeepSurv) and Deep Accelerated Failure Time (DeepAFT) models. We additionally provide theoretical support for the proposed DeepEH model by establishing consistency and convergence rate of the survival function estimator, which underscore the attractive feature that deep learning is able to detect low-dimensional structure of data in high-dimensional space. Numerical experiments also provide evidence that the proposed methods outperform existing statistical and deep learning approaches to survival analysis.

The inaccuracy of neural network models on inputs that do not stem from the distribution underlying the training data is problematic and at times unrecognized. Uncertainty estimates of model predictions are often based on the variation in predictions produced by a diverse ensemble of models applied to the same input. Here we describe Maximize Overall Diversity (MOD), an approach to improve ensemble-based uncertainty estimates by encouraging larger overall diversity in ensemble predictions across all possible inputs. We apply MOD to regression tasks including 38 Protein-DNA binding datasets, 9 UCI datasets, and the IMDB-Wiki image dataset. We also explore variants that utilize adversarial training techniques and data density estimation. For out-of-distribution test examples, MOD significantly improves predictive performance and uncertainty calibration without sacrificing performance on test data drawn from same distribution as the training data. We also find that in Bayesian optimization tasks, the performance of UCB acquisition is improved via MOD uncertainty estimates.

Generative autoencoders offer a promising approach for controllable text generation by leveraging their learned sentence representations. However, current models struggle to maintain coherent latent spaces required to perform meaningful text manipulations via latent vector operations. Specifically, we demonstrate by example that neural encoders do not necessarily map similar sentences to nearby latent vectors. A theoretical explanation for this phenomenon establishes that high-capacity autoencoders can learn an arbitrary mapping between sequences and associated latent representations. To remedy this issue, we augment adversarial autoencoders with a denoising objective where original sentences are reconstructed from perturbed versions (referred to as DAAE). We prove that this simple modification guides the latent space geometry of the resulting model by encouraging the encoder to map similar texts to similar latent representations. In empirical comparisons with various types of autoencoders, our model provides the best trade-off between generation quality and reconstruction capacity. Moreover, the improved geometry of the DAAE latent space enables \emph{zero-shot} text style transfer via simple latent vector arithmetic.

Automated machine learning (AutoML) offers the promise of translating raw data into accurate predictions with just a few lines of code. Rather than relying on human time/effort and manual experimentation, models can be improved by simply letting the AutoML system run for more time. In this hands-on tutorial, we demonstrate fundamental techniques that enable powerful AutoML. We consider standard supervised learning tasks on various types of data including tables, text, images, as well as multi-modal data comprised of multiple types. Rather than technical descriptions of how individual ML models work, we emphasize how to best use models within an overall ML pipeline that takes in raw training data and outputs pre-dictions for test data. A major focus of our tutorial is on automating deep learning, a class of powerful techniques that are cumbersome to manage manually. Despite this, hardly any educational material describes their successful automation. Each topic covered in the tutorial is accompanied by a hands-on Jupyter notebook that implements best practices (which will be available on Github before and after the tutorial). Most of this code is adopted from AutoGluon (autogluon.mxnet.io), a recent AutoML toolkit for automated deep learning that is both state-of-the-art and easy-to-use.

Automated machine learning (AutoML) can produce complex model ensembles by stacking, bagging, and boosting many individual models like trees, deep networks, and nearest neighbor estimators. While highly accurate, the resulting predictors are large, slow, and opaque as compared to their constituents. To improve the deployment of AutoML on tabular data, we propose FAST-DAD to distill arbitrarily-complex ensemble predictors into individual models like boosted trees, random forests, and deep networks. At the heart of our approach is a data augmentation strategy based on Gibbs sampling from a self-attention pseudolikelihood estimator. Across 30 datasets spanning regression and binary/multiclass classification tasks, FAST-DAD distillation produces significantly better individual models than one obtains through standard training on the original data. Our individual distilled models are over 10x faster and more accurate than ensemble predictors produced by AutoML tools like H2O/AutoSklearn.

MOTIVATION: The precise targeting of antibodies and other protein therapeutics is required for their proper function and the elimination of deleterious off-target effects. Often the molecular structure of a therapeutic target is unknown and randomized methods are used to design antibodies without a model that relates antibody sequence to desired properties. RESULTS: Here, we present Ens-Grad, a machine learning method that can design complementarity determining regions of human Immunoglobulin G antibodies with target affinities that are superior to candidates derived from phage display panning experiments. We also demonstrate that machine learning can improve target specificity by the modular composition of models from different experimental campaigns, enabling a new integrative approach to improving target specificity. Our results suggest a new path for the discovery of therapeutic molecules by demonstrating that predictive and differentiable models of antibody binding can be learned from high-throughput experimental data without the need for target structural data. AVAILABILITY AND IMPLEMENTATION: Sequencing data of the phage panning experiment are deposited at NIH's Sequence Read Archive (SRA) under the accession number SRP158510. We make our code available at https://github.com/gifford-lab/antibody-2019. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Local explanation frameworks aim to rationalize particular decisions made by a black-box prediction model. Existing techniques are often restricted to a specific type of predictor or based on input saliency, which may be undesirably sensitive to factors unrelated to the model’s decision making process. We instead propose sufficient input subsets that identify minimal subsets of features whose observed values alone suffice for the same decision to be reached, even if all other input feature values are missing. General principles that globally govern a model’s decision-making can also be revealed by searching for clusters of such input patterns across many data points. Our approach is conceptually straightforward, entirely model-agnostic, simply implemented using instance-wise backward selection, and able to produce more concise rationales than existing techniques. We demonstrate the utility of our interpretation method on various neural network models trained on text, image, and genomic data.

Zhijing Jin, Di Jin, Jonas Mueller, Nicholas Matthews, Enrico Santus. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

A key obstacle in automated analytics and meta-learning is failing to recognize when different datasets contain measurements of the same variable. Because provided attribute labels are often uninformative in practice, this task may be more robustly addressed by leveraging the data values themselves, rather than relying on their arbitrarily selected variable names. Here, we present a computationally efficient method to identify high-confidence variable matches between a given set of data values and a large repository of previously encountered datasets. Our approach enjoys numerous advantages over distributional similarity based techniques because we leverage learned vector embeddings of datasets which adaptively account for natural forms of data variation encountered in practice. Based on the neural architecture of deep sets, our embeddings can be computed for both numeric and string data, and empirically outperform standard statistical techniques for dataset search.

We consider dynamic pricing with many products under an evolving but low-dimensional demand model. Assuming the temporal variation in cross-elasticities exhibits low-rank structure based on fixed (latent) features of the products, we show that the revenue maximization problem reduces to an online bandit convex optimization with side information given by the observed demands. We design dynamic pricing algorithms whose revenue approaches that of the best fixed price vector in hindsight, at a rate that only depends on the intrinsic rank of the demand model and not the number of products. Our approach applies a bandit convex optimization algorithm in a projected low-dimensional space spanned by the latent product features, while simultaneously learning this span via online singular value decomposition of a carefully-crafted matrix containing the observed demands.

Our goal is to identify beneficial interventions from observational data. We consider interventions that are narrowly focused (impacting few covariates) and may be tailored to each individual or globally enacted over a population. For applications where harmful intervention is drastically worse than proposing no change, we propose a conservative definition of the optimal intervention. Assuming the underlying relationship remains invariant under intervention, we develop efficient algorithms to identify the optimal intervention policy from limited data and provide theoretical guarantees for our approach in a Gaussian Process setting. Although our methods assume covariates can be precisely adjusted, they remain capable of improving outcomes in misspecified settings with unintentional downstream effects. Empirically, our approach identifies good interventions in two practical applications: gene perturbation and writing improvement.

We present a model that, after learning on observations of (sequence, outcome) pairs, can be efficiently used to revise a new sequence in order to improve its associated outcome. Our framework requires neither example improvements, nor additional evaluation of outcomes for proposed revisions. To avoid combinatorial-search over sequence elements, we specify a generative model with continuous latent factors, which is learned via joint approximate inference using a recurrent variational autoencoder (VAE) and an outcome-predicting neural network module. Under this model, gradient methods can be used to efficiently optimize the continuous latent factors with respect to inferred outcomes. By appropriately constraining this optimization and using the VAE decoder to generate a revised sequence, we ensure the revision is fundamentally similar to the original sequence, is associated with better outcomes, and looks natural. These desiderata are proven to hold with high probability under our approach, which is empirically demonstrated for revising natural language sentences.

We present a siamese adaptation of the Long Short-Term Memory (LSTM) network for labeled data comprised of pairs of variable-length sequences. Our model is applied to assess semantic similarity between sentences, where we exceed state of the art, outperforming carefully handcrafted features and recently proposed neural network systems of greater complexity. For these applications, we provide word-embedding vectors supplemented with synonymic information to the LSTMs, which use a fixed size vector to encode the underlying meaning expressed in a sentence (irrespective of the particular wording/syntax). By restricting subsequent operations to rely on a simple Manhattan metric, we compel the sentence representations learned by our model to form a highly structured space whose geometry reflects complex semantic relationships. Our results are the latest in a line of findings that showcase LSTMs as powerful language models capable of tasks requiring intricate understanding.

We introduce principal differences analysis for analyzing differences between high-dimensional distributions. The method operates by finding the projection that maximizes the Wasserstein divergence between the resulting univariate populations. Relying on the Cramer-Wold device, it requires no assumptions about the form of the underlying distributions, nor the nature of their inter-class differences. A sparse variant of the method is introduced to identify features responsible for the differences. We provide algorithms for both the original minimax formulation as well as its semidefinite relaxation. In addition to deriving some convergence results, we illustrate how the approach may be applied to identify differences between cell populations in the somatosensory cortex and hippocampus as manifested by single cell RNA-seq. Our broader framework extends beyond the specific choice of Wasserstein divergence.