Colorectal cancer (CRC) and gastric cancer (GC) are leading causes of cancer-related mortality. However, cost-effective methods for simultaneous detection of CRC and GC in circulating cell-free DNA (cfDNA) remain insufficiently explored. To address this, we developed targeted methylated CpG tandem amplification and sequencing (tMCTA-seq), a PCR-based method utilizing a set of locus-specific primers with a universal CGCGCGG primer, and targeted a panel of 110 loci. The method demonstrated high technical sensitivity below one haploid genome equivalent. Using a repeated nested cross-validation framework, the ensemble model, applied to 448 plasma samples (170 CRC, 101 GC, and 177 control participants), achieved areas under the curve (AUCs) of 0.928 (88.2% sensitivity and 90.7% specificity) for CRC and 0.926 (86.7% sensitivity and 94.4% specificity) for GC on the test set. Furthermore, tMCTA-seq differentiated between CRC and GC (AUC = 0.819). Thus, tMCTA-seq is a cost-effective, methylation-based approach for simultaneous detection of and differentiation between two major gastrointestinal cancers in blood.

Reward models (RMs) play a pivotal role in aligning large language models (LLMs) with human preferences. However, traditional RM training, which relies on response pairs tied to specific prompts, struggles to disentangle prompt-driven preferences from prompt-independent artifacts, such as response length and format. In this work, we expose a fundamental limitation of current RM training methods, where RMs fail to effectively distinguish between contextual signals and irrelevant artifacts when determining preferences. To address this, we introduce a causal framework that learns preferences independent of these artifacts and propose a novel data augmentation technique designed to eliminate them. Extensive experiments show that our approach successfully filters out undesirable artifacts, yielding a more robust reward model (RRM). Our RRM improves the performance of a pairwise reward model trained on Gemma-2-9b-it, on Reward-Bench, increasing accuracy from 80.61% to 84.15%. Additionally, we train two DPO policies using both the RM and RRM, demonstrating that the RRM significantly enhances DPO-aligned policies, improving MT-Bench scores from 7.27 to 8.31 and length-controlled win-rates in AlpacaEval-2 from 33.46% to 52.49%.

Modern deep models for summarization attains impressive benchmark performance, but they are prone to generating miscalibrated predictive uncertainty. This means that they assign high confidence to low-quality predictions, leading to compromised reliability and trustworthiness in real-world applications. Probabilistic deep learning methods are common solutions to the miscalibration problem. However, their relative effectiveness in complex autoregressive summarization tasks are not well-understood. In this work, we thoroughly investigate different state-of-the-art probabilistic methods' effectiveness in improving the uncertainty quality of the neural summarization models, across three large-scale benchmarks with varying difficulty using our newly introduced evaluation protocol. We show that the probabilistic methods consistently improve the model's generation and uncertainty quality, leading to improved selective generation performance (i.e., abstaining from low-quality summaries) in practice. We also reveal notable failure patterns of probabilistic methods widely-adopted in NLP community (e.g., Deep Ensemble and Monte Carlo Dropout), cautioning the importance of choosing appropriate method for the data setting.

We introduce a new deep generative model useful for uncertainty quantification: the Morse neural network, which generalizes the unnormalized Gaussian densities to have modes of high-dimensional submanifolds instead of just discrete points. Fitting the Morse neural network via a KL-divergence loss yields 1) a (unnormalized) generative density, 2) an OOD detector, 3) a calibration temperature, 4) a generative sampler, along with in the supervised case 5) a distance aware-classifier. The Morse network can be used on top of a pre-trained network to bring distance-aware calibration w.r.t the training data. Because of its versatility, the Morse neural networks unifies many techniques: e.g., the Entropic Out-of-Distribution Detector of (Macêdo et al., 2021) in OOD detection, the one class Deep Support Vector Description method of (Ruff et al., 2018) in anomaly detection, or the Contrastive One Class classifier in continuous learning (Sun et al., 2021). The Morse neural network has connections to support vector machines, kernel methods, and Morse theory in topology.

Contrastively trained text-image models have the remarkable ability to perform zero-shot classification, that is, classifying previously unseen images into categories that the model has never been explicitly trained to identify. However, these zero-shot classifiers need prompt engineering to achieve high accuracy. Prompt engineering typically requires hand-crafting a set of prompts for individual downstream tasks. In this work, we aim to automate this prompt engineering and improve zero-shot accuracy through prompt ensembling. In particular, we ask "Given a large pool of prompts, can we automatically score the prompts and ensemble those that are most suitable for a particular downstream dataset, without needing access to labeled validation data?". We demonstrate that this is possible. In doing so, we identify several pathologies in a naive prompt scoring method where the score can be easily overconfident due to biases in pre-training and test data, and we propose a novel prompt scoring method that corrects for the biases. Using our proposed scoring method to create a weighted average prompt ensemble, our method outperforms equal average ensemble, as well as hand-crafted prompts, on ImageNet, 4 of its variants, and 11 fine-grained classification benchmarks, all while being fully automatic, optimization-free, and not requiring access to labeled validation data.

Safe deployment of large language models (LLMs) may benefit from a reliable method for assessing their generated content to determine when to abstain or to selectively generate. While likelihood-based metrics such as perplexity are widely employed, recent research has demonstrated the limitations of using sequence-level probability estimates given by LLMs as reliable indicators of generation quality. Conversely, LLMs have demonstrated strong calibration at the token level, particularly when it comes to choosing correct answers in multiple-choice questions or evaluating true/false statements. In this work, we reformulate open-ended generation tasks into token-level prediction tasks, and leverage LLMs' superior calibration at the token level. We instruct an LLM to self-evaluate its answers, employing either a multi-way comparison or a point-wise evaluation approach, with the option to include a ``None of the above'' option to express the model's uncertainty explicitly. We benchmark a range of scoring methods based on self-evaluation and evaluate their performance in selective generation using TruthfulQA and TL;DR. Through experiments with PaLM-2 and GPT-3, we demonstrate that self-evaluation based scores not only improve accuracy, but also correlate better with the overall quality of generated content.

Multi-modal- image-text models such as CLIP and LiT have demonstrated impressive performance on image classification benchmarks and their zero-shot generalization ability is particularly exciting. While the top-5 zero-shot accuracies of these models are very high, the top-1 accuracies are much lower (over 25% gap in some cases). We investigate the reasons for this performance gap and find that many of the failure cases are caused by ambiguity in the text prompts. First, we develop a simple and efficient zero-shot post-hoc method to identify images whose top-1 prediction is likely to be incorrect, by measuring consistency of the predictions w.r.t. multiple prompts and image transformations. We show that our procedure better predicts mistakes, outperforming the popular max logit baseline on selective prediction tasks. Next, we propose a simple and efficient way to improve accuracy on such uncertain images by making use of the WordNet hierarchy; specifically we augment the original class by incorporating its parent and children from the semantic label hierarchy, and plug the augmentation into text prompts. We conduct experiments on both CLIP and LiT models with five different ImageNet-based datasets. For CLIP, our method improves the top-1 accuracy by 17.13% on the uncertain subset and 3.6% on the entire ImageNet validation set. We also show that our method improves across ImageNet shifted datasets, four other datasets, and other model architectures such as LiT. The proposed method 1 1 Work carried out mainly at Google is hyperparameter-free, requires no additional model training and can be easily scaled to other large multi-modal architectures. Code is available at https://github.com/gyhandy/Hierarchy-CLIP.

We focus on the challenge of out-of-distribution (OOD) detection in deep learning models, a crucial aspect in ensuring reliability. Despite considerable effort, the problem remains significantly challenging in deep learning models due to their propensity to output over-confident predictions for OOD inputs. We propose a novel one-class open-set OOD detector that leverages text-image pre-trained models in a zero-shot fashion and incorporates various descriptions of in-domain and OOD. Our approach is designed to detect anything not in-domain and offers the flexibility to detect a wide variety of OOD, defined via fine- or coarse-grained labels, or even in natural language. We evaluate our approach on challenging benchmarks including large-scale datasets containing fine-grained, semantically similar classes, distributionally shifted images, and multi-object images containing a mixture of in-domain and OOD objects. Our method shows superior performance over previous methods on all benchmarks. Code is available at https://github.com/gyhandy/One-Class-Anything

The evaluation of abstractive summarization models typically uses test data that is identically distributed as training data. In real-world practice, documents to be summarized may contain input noise caused by text extraction artifacts or data pipeline bugs. The robustness of model performance under distribution shift caused by such noise is relatively under-studied. We present a large empirical study quantifying the sometimes severe loss in performance (up to 12 ROUGE-1 points) from different types of input noise for a range of datasets and model sizes. We then propose a light-weight method for detecting and removing such noise in the input during model inference without requiring any extra training, auxiliary models, or even prior knowledge of the type of noise. Our proposed approach effectively mitigates the loss in performance, recovering a large fraction of the performance drop, sometimes as large as 11 ROUGE-1 points.

Machine learning or deep learning models have been widely used for taxonomic classification of metagenomic sequences and many studies reported high classification accuracy. Such models are usually trained based on sequences in several training classes in hope of accurately classifying unknown sequences into these classes. However, when deploying the classification models on real testing data sets, sequences that do not belong to any of the training classes may be present and are falsely assigned to one of the training classes with high confidence. Such sequences are referred to as out-of-distribution (OOD) sequences and are ubiquitous in metagenomic studies. To address this problem, we develop a deep generative model-based method, MLR-OOD, that measures the probability of a testing sequencing belonging to OOD by the likelihood ratio of the maximum of the in-distribution (ID) class conditional likelihoods and the Markov chain likelihood of the testing sequence measuring the sequence complexity. We compose three different microbial data sets consisting of bacterial, viral, and plasmid sequences for comprehensively benchmarking OOD detection methods. We show that MLR-OOD achieves the state-of-the-art performance demonstrating the generality of MLR-OOD to various types of microbial data sets. It is also shown that MLR-OOD is robust to the GC content, which is a major confounding effect for OOD detection of genomic sequences. In conclusion, MLR-OOD will greatly reduce false positives caused by OOD sequences in metagenomic sequence classification.

The statistical inference of high-order Markov chains (MCs) for biological sequences is vital for molecular sequence analyses but can be hindered by the high dimensionality of free parameters. In the seminal article by Bühlmann and Wyner, variable length Markov chain (VLMC) model was proposed to embed the full-order MC in a sparse structured context tree. In the key procedure of tree pruning of their proposed context algorithm, the word count-based statistic for each branch was defined and compared with a fixed cutoff threshold calculated from a common chi-square distribution to prune the branch of the context tree. In this study, we find that the word counts for each branch are highly intercorrelated, resulting in non-negligible effects on the distribution of the statistic of interest. We demonstrate that the inferred context tree based on the original context algorithm by Bühlmann and Wyner, which uses a fixed cutoff threshold based on a common chi-square distribution, can be systematically biased and error prone. We denote the original context algorithm as VLMC-Biased (VLMC-B). To solve this problem, we propose a new context tree inference algorithm using an adaptive tree-pruning scheme, termed VLMC-Consistent (VLMC-C). The VLMC-C is founded on the consistent branch-specific mixed chi-square distributions calculated based on asymptotic normal distribution of multiple word patterns. We validate our theoretical branch-specific asymptotic distribution using simulated data. We compare VLMC-C with VLMC-B on context tree inference using both simulated and real genome sequence data and demonstrate that VLMC-C outperforms VLMC-B for both context tree reconstruction accuracy and model compression capacity.

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.

Accurate uncertainty quantification is a major challenge in deep learning, as neural networks can make overconfident errors and assign high confidence predictions to out-of-distribution (OOD) inputs. The most popular approaches to estimate predictive uncertainty in deep learning are methods that combine predictions from multiple neural networks, such as Bayesian neural networks (BNNs) and deep ensembles. However their practicality in real-time, industrial-scale applications are limited due to the high memory and computational cost. Furthermore, ensembles and BNNs do not necessarily fix all the issues with the underlying member networks. In this work, we study principled approaches to improve uncertainty property of a single network, based on a single, deterministic representation. By formalizing the uncertainty quantification as a minimax learning problem, we first identify distance awareness, i.e., the model's ability to quantify the distance of a testing example from the training data, as a necessary condition for a DNN to achieve high-quality (i.e., minimax optimal) uncertainty estimation. We then propose Spectral-normalized Neural Gaussian Process (SNGP), a simple method that improves the distance-awareness ability of modern DNNs with two simple changes: (1) applying spectral normalization to hidden weights to enforce bi-Lipschitz smoothness in representations and (2) replacing the last output layer with a Gaussian process layer. On a suite of vision and language understanding benchmarks, SNGP outperforms other single-model approaches in prediction, calibration and out-of-domain detection. Furthermore, SNGP provides complementary benefits to popular techniques such as deep ensembles and data augmentation, making it a simple and scalable building block for probabilistic deep learning. Code is open-sourced at https://github.com/google/uncertainty-baselines

We develop and rigorously evaluate a deep learning based system that can accurately classify skin conditions while detecting rare conditions for which there is not enough data available for training a confident classifier. We frame this task as an out-of-distribution (OOD) detection problem. Our novel approach, hierarchical outlier detection (HOD) assigns multiple abstention classes for each training outlier class and jointly performs a coarse classification of inliers vs. outliers, along with fine-grained classification of the individual classes. We demonstrate the effectiveness of the HOD loss in conjunction with modern representation learning approaches (BiT, SimCLR, MICLe) and explore different ensembling strategies for further improving the results. We perform an extensive subgroup analysis over conditions of varying risk levels and different skin types to investigate how the OOD detection performance changes over each subgroup and demonstrate the gains of our framework in comparison to baselines. Finally, we introduce a cost metric to approximate downstream clinical impact. We use this cost metric to compare the proposed method against a baseline system, thereby making a stronger case for the overall system effectiveness in a real-world deployment scenario.

Mahalanobis distance (MD) is a simple and popular post-processing method for detecting out-of-distribution (OOD) inputs in neural networks. We analyze its failure modes for near-OOD detection and propose a simple fix called relative Mahalanobis distance (RMD) which improves performance and is more robust to hyperparameter choice. On a wide selection of challenging vision, language, and biology OOD benchmarks (CIFAR-100 vs CIFAR-10, CLINC OOD intent detection, Genomics OOD), we show that RMD meaningfully improves upon MD performance (by up to 15% AUROC on genomics OOD).

MOTIVATION:The rapid development of sequencing technologies has enabled us to generate a large number of metagenomic reads from genetic materials in microbial communities, making it possible to gain deep insights into understanding the differences between the genetic materials of different groups of microorganisms, such as bacteria, viruses, plasmids, etc. Computational methods based on k-mer frequencies have been shown to be highly effective for classifying metagenomic sequencing reads into different groups. However, such methods usually use all the k-mers as features for prediction without selecting relevant k-mers for the different groups of sequences, i.e. unique nucleotide patterns containing biological significance. RESULTS:To select k-mers for distinguishing different groups of sequences with guaranteed false discovery rate (FDR) control, we develop KIMI, a general framework based on model-X Knockoffs regarded as the state-of-the-art statistical method for FDR control, for sequence motif discovery with arbitrary target FDR level, such that reproducibility can be theoretically guaranteed. KIMI is shown through simulation studies to be effective in simultaneously controlling FDR and yielding high power, outperforming the broadly used Benjamini-Hochberg procedure and the q-value method for FDR control. To illustrate the usefulness of KIMI in analyzing real datasets, we take the viral motif discovery problem as an example and implement KIMI on a real dataset consisting of viral and bacterial contigs. We show that the accuracy of predicting viral and bacterial contigs can be increased by training the prediction model only on relevant k-mers selected by KIMI. AVAILABILITYAND IMPLEMENTATION:Our implementation of KIMI is available at https://github.com/xinbaiusc/KIMI. SUPPLEMENTARY INFORMATION:Supplementary data are available at Bioinformatics online.

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.

Near out-of-distribution detection (OOD) is a major challenge for deep neural networks. We demonstrate that large-scale pre-trained transformers can significantly improve the state-of-the-art (SOTA) on a range of near OOD tasks across different data modalities. For instance, on CIFAR-100 vs CIFAR-10 OOD detection, we improve the AUROC from 85% (current SOTA) to more than 96% using Vision Transformers pre-trained on ImageNet-21k. On a challenging genomics OOD detection benchmark, we improve the AUROC from 66% to 77% using transformers and unsupervised pre-training. To further improve performance, we explore the few-shot outlier exposure setting where a few examples from outlier classes may be available; we show that pre-trained transformers are particularly well-suited for outlier exposure, and that the AUROC of OOD detection on CIFAR-100 vs CIFAR-10 can be improved to 98.7% with just 1 image per OOD class, and 99.46% with 10 images per OOD class. For multi-modal image-text pre-trained transformers such as CLIP, we explore a new way of using just the names of outlier classes as a sole source of information without any accompanying images, and show that this outperforms previous SOTA on standard vision OOD benchmark tasks.

For many types of high-throughput sequencing experiments, success in downstream analysis depends on attaining sufficient coverage for individual positions in the genome. For example, when identifying single-nucleotide variants de novo, the number of reads supporting a particular variant call determines our confidence in that variant call. If sequenced reads are distributed uniformly along the genome, the coverage of a nucleotide position is easily approximated by a Poisson distribution, with rate equal to average sequencing depth. Unfortunately, as has become well known, high-throughput sequencing data are never uniform. The numerous factors contributing to variation in coverage have resisted attempts at direct modeling and change along with minor adjustments in the underlying technology. We propose a new nonparametric method to predict the portion of a genome that will attain some specified minimum coverage, as a function of sequencing effort, using information from a shallow sequencing experiment from the same library. Simulations show our approach performs well under an array of distributional assumptions that deviate from uniformity. We applied this approach to estimate coverage at varying depths in single-cell whole-genome sequencing data from multiple protocols. These resulted in highly accurate predictions, demonstrating the effectiveness of our approach in analyzing complexity of sequencing libraries and optimizing design of sequencing experiments.

Background: Markov chains (MC) have been widely used to model molecular sequences. The estimations of MC transition matrix and confidence intervals of the transition probabilities from long sequence data have been intensively studied in the past decades. In next generation sequencing (NGS), a large amount of short reads are generated. These short reads can overlap and some regions of the genome may not be sequenced resulting in a new type of data. Based on NGS data, the transition probabilities of MC can be estimated by moment estimators. However, the classical asymptotic distribution theory for MC transition probability estimators based on long sequences is no longer valid.Methods: In this study, we present the asymptotic distributions of several statistics related to MC based on NGS data. We show that, after scaling by the effective coverage d defined in a previous study by the authors, these statistics based on NGS data approximate to the same distributions as the corresponding statistics for long sequences.Results: We apply the asymptotic properties of these statistics for finding the theoretical confidence regions for MC transition probabilities based on NGS short reads data. We validate our theoretical confidence intervals using both simulated data and real data sets, and compare the results with those by the parametric bootstrap method.Conclusions: We find that the asymptotic distributions of these statistics and the theoretical confidence intervals of transition probabilities based on NGS data given in this study are highly accurate, providing a powerful tool for NGS data analysis.