Abstract
We show that the effectiveness of the well celebrated Mixup [Zhang et al., 2018] can be further improved if instead of using it as the sole learning objective, it is utilized as an additional regularizer to the standard cross-entropy loss. This simple change not only provides much improved accuracy but also significantly improves the quality of the predictive uncertainty estimation of Mixup in most cases under various forms of covariate shifts and out-of-distribution detection experiments. In fact, we observe that Mixup yields much degraded performance on detecting out-of-distribution samples possibly, as we show empirically, because of its tendency to learn models that exhibit high-entropy throughout; making it difficult to differentiate in-distribution samples from out-distribution ones. To show the efficacy of our approach (RegMixup), we provide thorough analyses and experiments on vision datasets (ImageNet & CIFAR-10/100) and compare it with a suite of recent approaches for reliable uncertainty estimation.
Keyword: expected calibration error
Uncertainty-aware Panoptic Segmentation
Authors: Kshitij Sirohi, Sajad Marvi, Daniel Büscher, Wolfram Burgard
Subjects: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
Abstract
Reliable scene understanding is indispensable for modern autonomous systems. Current learning-based methods typically try to maximize their performance based on segmentation metrics that only consider the quality of the segmentation. However, for the safe operation of a system in the real world it is crucial to consider the uncertainty in the prediction as well. In this work, we introduce the novel task of uncertainty-aware panoptic segmentation, which aims to predict per-pixel semantic and instance segmentations, together with per-pixel uncertainty estimates. We define two novel metrics to facilitate its quantitative analysis, the uncertainty-aware Panoptic Quality (uPQ) and the panoptic Expected Calibration Error (pECE). We further propose the novel top-down Evidential Panoptic Segmentation Network (EvPSNet) to solve this task. Our architecture employs a simple yet effective probabilistic fusion module that leverages the predicted uncertainties. Additionally, we propose a new Lov\'asz evidential loss function to optimize the IoU for the segmentation utilizing the probabilities provided by deep evidential learning. Furthermore, we provide several strong baselines combining state-of-the-art panoptic segmentation networks with sampling-free uncertainty estimation techniques. Extensive evaluations show that our EvPSNet achieves the new state-of-the-art for the standard Panoptic Quality (PQ), as well as for our uncertainty-aware panoptic metrics.
Keyword: overconfident
There is no result
Keyword: overconfidence
There is no result
Keyword: confidence
Semi-supervised Contrastive Outlier removal for Pseudo Expectation Maximization (SCOPE)
Abstract
Semi-supervised learning is the problem of training an accurate predictive model by combining a small labeled dataset with a presumably much larger unlabeled dataset. Many methods for semi-supervised deep learning have been developed, including pseudolabeling, consistency regularization, and contrastive learning techniques. Pseudolabeling methods however are highly susceptible to confounding, in which erroneous pseudolabels are assumed to be true labels in early iterations, thereby causing the model to reinforce its prior biases and thereby fail to generalize to strong predictive performance. We present a new approach to suppress confounding errors through a method we describe as Semi-supervised Contrastive Outlier removal for Pseudo Expectation Maximization (SCOPE). Like basic pseudolabeling, SCOPE is related to Expectation Maximization (EM), a latent variable framework which can be extended toward understanding cluster-assumption deep semi-supervised algorithms. However, unlike basic pseudolabeling which fails to adequately take into account the probability of the unlabeled samples given the model, SCOPE introduces an outlier suppression term designed to improve the behavior of EM iteration given a discrimination DNN backbone in the presence of outliers. Our results show that SCOPE greatly improves semi-supervised classification accuracy over a baseline, and furthermore when combined with consistency regularization achieves the highest reported accuracy for the semi-supervised CIFAR-10 classification task using 250 and 4000 labeled samples. Moreover, we show that SCOPE reduces the prevalence of confounding errors during pseudolabeling iterations by pruning erroneous high-confidence pseudolabeled samples that would otherwise contaminate the labeled set in subsequent retraining iterations.
Collecting high-quality adversarial data for machine reading comprehension tasks with humans and models in the loop
Authors: Damian Y. Romero Diaz, Magdalena Anioł, John Culnan
Subjects: Computation and Language (cs.CL); Human-Computer Interaction (cs.HC); Machine Learning (cs.LG)
Abstract
We present our experience as annotators in the creation of high-quality, adversarial machine-reading-comprehension data for extractive QA for Task 1 of the First Workshop on Dynamic Adversarial Data Collection (DADC). DADC is an emergent data collection paradigm with both models and humans in the loop. We set up a quasi-experimental annotation design and perform quantitative analyses across groups with different numbers of annotators focusing on successful adversarial attacks, cost analysis, and annotator confidence correlation. We further perform a qualitative analysis of our perceived difficulty of the task given the different topics of the passages in our dataset and conclude with recommendations and suggestions that might be of value to people working on future DADC tasks and related annotation interfaces.
What Can Secondary Predictions Tell Us? An Exploration on Question-Answering with SQuAD-v2.0
Authors: Michael Kamfonas, Gabriel Alon
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Abstract
Performance in natural language processing, and specifically for the question-answer task, is typically measured by comparing a model\'s most confident (primary) prediction to golden answers (the ground truth). We are making the case that it is also useful to quantify how close a model came to predicting a correct answer even for examples that failed. We define the Golden Rank (GR) of an example as the rank of its most confident prediction that exactly matches a ground truth, and show why such a match always exists. For the 16 transformer models we analyzed, the majority of exactly matched golden answers in secondary prediction space hover very close to the top rank. We refer to secondary predictions as those ranking above 0 in descending confidence probability order. We demonstrate how the GR can be used to classify questions and visualize their spectrum of difficulty, from persistent near successes to persistent extreme failures. We derive a new aggregate statistic over entire test sets, named the Golden Rank Interpolated Median (GRIM) that quantifies the proximity of failed predictions to the top choice made by the model. To develop some intuition and explore the applicability of these metrics we use the Stanford Question Answering Dataset (SQuAD-2) and a few popular transformer models from the Hugging Face hub. We first demonstrate that the GRIM is not directly correlated with the F1 and exact match (EM) scores. We then calculate and visualize these scores for various transformer architectures, probe their applicability in error analysis by clustering failed predictions, and compare how they relate to other training diagnostics such as the EM and F1 scores. We finally suggest various research goals, such as broadening data collection for these metrics and their possible use in adversarial training.
Constructing MDP Abstractions Using Data with Formal Guarantees
Authors: Abolfazl Lavaei, Sadegh Soudjani, Emilio Frazzoli, Majid Zamani
Abstract
This paper is concerned with a data-driven technique for constructing finite Markov decision processes (MDPs) as finite abstractions of discrete-time stochastic control systems with unknown dynamics while providing formal closeness guarantees. The proposed scheme is based on notions of stochastic bisimulation functions (SBF) to capture the probabilistic distance between state trajectories of an unknown stochastic system and those of finite MDP. In our proposed setting, we first reformulate corresponding conditions of SBF as a robust convex program (RCP). We then propose a scenario convex program (SCP) associated to the original RCP by collecting a finite number of data from trajectories of the system. We ultimately construct an SBF between the data-driven finite MDP and the unknown stochastic system with a given confidence level by establishing a probabilistic relation between optimal values of the SCP and the RCP. We also propose two different approaches for the construction of finite MDPs from data. We illustrate the efficacy of our results over a nonlinear jet engine compressor with unknown dynamics. We construct a data-driven finite MDP as a suitable substitute of the original system to synthesize controllers maintaining the system in a safe set with some probability of satisfaction and a desirable confidence level.
Mining Seasonal Temporal Patterns in Big Time Series
Authors: Van Long Ho, Nguyen Ho, Torben Bach Pedersen
Abstract
Very large time series are increasingly available from an ever wider range of IoT-enabled sensors, from which significant insights can be obtained through mining temporal patterns from them. A useful type of patterns found in many real-world applications exhibits periodic occurrences, and is thus called seasonal temporal pattern (STP). Compared to regular patterns, mining seasonal temporal patterns is more challenging since traditional measures such as support and confidence do not capture the seasonality characteristics. Further, the anti-monotonicity property does not hold for STPs, and thus, resulting in an exponential search space. This paper presents our Frequent Seasonal Temporal Pattern Mining from Time Series (FreqSTPfTS) solution providing: (1) The first solution for seasonal temporal pattern mining (STPM) from time series that can mine STP at different data granularities. (2) The STPM algorithm that uses efficient data structures and two pruning techniques to reduce the search space and speed up the mining process. (3) An approximate version of STPM that uses mutual information, a measure of data correlation, to prune unpromising time series from the search space. (4) An extensive experimental evaluation showing that STPM outperforms the baseline in runtime and memory consumption, and can scale to big datasets. The approximate STPM is up to an order of magnitude faster and less memory consuming than the baseline, while maintaining high accuracy.
Two-Stage Neural Contextual Bandits for Personalised News Recommendation
Abstract
We consider the problem of personalised news recommendation where each user consumes news in a sequential fashion. Existing personalised news recommendation methods focus on exploiting user interests and ignores exploration in recommendation, which leads to biased feedback loops and hurt recommendation quality in the long term. We build on contextual bandits recommendation strategies which naturally address the exploitation-exploration trade-off. The main challenges are the computational efficiency for exploring the large-scale item space and utilising the deep representations with uncertainty. We propose a two-stage hierarchical topic-news deep contextual bandits framework to efficiently learn user preferences when there are many news items. We use deep learning representations for users and news, and generalise the neural upper confidence bound (UCB) policies to generalised additive UCB and bilinear UCB. Empirical results on a large-scale news recommendation dataset show that our proposed policies are efficient and outperform the baseline bandit policies.
Keyword: scaling
NumS: Scalable Array Programming for the Cloud
Authors: Melih Elibol, Vinamra Benara, Samyu Yagati, Lianmin Zheng, Alvin Cheung, Michael I. Jordan, Ion Stoica
Abstract
Scientists increasingly rely on Python tools to perform scalable distributed memory array operations using rich, NumPy-like expressions. However, many of these tools rely on dynamic schedulers optimized for abstract task graphs, which often encounter memory and network bandwidth-related bottlenecks due to sub-optimal data and operator placement decisions. Tools built on the message passing interface (MPI), such as ScaLAPACK and SLATE, have better scaling properties, but these solutions require specialized knowledge to use. In this work, we present NumS, an array programming library which optimizes NumPy-like expressions on task-based distributed systems. This is achieved through a novel scheduler called Load Simulated Hierarchical Scheduling (LSHS). LSHS is a local search method which optimizes operator placement by minimizing maximum memory and network load on any given node within a distributed system. Coupled with a heuristic for load balanced data layouts, our approach is capable of attaining communication lower bounds on some common numerical operations, and our empirical study shows that LSHS enhances performance on Ray by decreasing network load by a factor of 2x, requiring 4x less memory, and reducing execution time by 10x on the logistic regression problem. On terabyte-scale data, NumS achieves competitive performance to SLATE on DGEMM, up to 20x speedup over Dask on a key operation for tensor factorization, and a 2x speedup on logistic regression compared to Dask ML and Spark's MLlib.
Left Heavy Tails and the Effectiveness of the Policy and Value Networks in DNN-based best-first search for Sokoban Planning
Abstract
Despite the success of practical solvers in various NP-complete domains such as SAT and CSP as well as using deep reinforcement learning to tackle two-player games such as Go, certain classes of PSPACE-hard planning problems have remained out of reach. Even carefully designed domain-specialized solvers can fail quickly due to the exponential search space on hard instances. Recent works that combine traditional search methods, such as best-first search and Monte Carlo tree search, with Deep Neural Networks' (DNN) heuristics have shown promising progress and can solve a significant number of hard planning instances beyond specialized solvers. To better understand why these approaches work, we studied the interplay of the policy and value networks of DNN-based best-first search on Sokoban and show the surprising effectiveness of the policy network, further enhanced by the value network, as a guiding heuristic for the search. To further understand the phenomena, we studied the cost distribution of the search algorithms and found that Sokoban instances can have heavy-tailed runtime distributions, with tails both on the left and right-hand sides. In particular, for the first time, we show the existence of \textit{left heavy tails} and propose an abstract tree model that can empirically explain the appearance of these tails. The experiments show the critical role of the policy network as a powerful heuristic guiding the search, which can lead to left heavy tails with polynomial scaling by avoiding exploring exponentially sized subtrees. Our results also demonstrate the importance of random restarts, as are widely used in traditional combinatorial solvers, for DNN-based search methods to avoid left and right heavy tails.
Beyond neural scaling laws: beating power law scaling via data pruning
Authors: Ben Sorscher, Robert Geirhos, Shashank Shekhar, Surya Ganguli, Ari S. Morcos
Abstract
Widely observed neural scaling laws, in which error falls off as a power of the training set size, model size, or both, have driven substantial performance improvements in deep learning. However, these improvements through scaling alone require considerable costs in compute and energy. Here we focus on the scaling of error with dataset size and show how both in theory and practice we can break beyond power law scaling and reduce it to exponential scaling instead if we have access to a high-quality data pruning metric that ranks the order in which training examples should be discarded to achieve any pruned dataset size. We then test this new exponential scaling prediction with pruned dataset size empirically, and indeed observe better than power law scaling performance on ResNets trained on CIFAR-10, SVHN, and ImageNet. Given the importance of finding high-quality pruning metrics, we perform the first large-scale benchmarking study of ten different data pruning metrics on ImageNet. We find most existing high performing metrics scale poorly to ImageNet, while the best are computationally intensive and require labels for every image. We therefore developed a new simple, cheap and scalable self-supervised pruning metric that demonstrates comparable performance to the best supervised metrics. Overall, our work suggests that the discovery of good data-pruning metrics may provide a viable path forward to substantially improved neural scaling laws, thereby reducing the resource costs of modern deep learning.
Keyword: calibration
EBMs vs. CL: Exploring Self-Supervised Visual Pretraining for Visual Question Answering
Authors: Violetta Shevchenko, Ehsan Abbasnejad, Anthony Dick, Anton van den Hengel, Damien Teney
Subjects: Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL); Machine Learning (cs.LG)
Abstract
The availability of clean and diverse labeled data is a major roadblock for training models on complex tasks such as visual question answering (VQA). The extensive work on large vision-and-language models has shown that self-supervised learning is effective for pretraining multimodal interactions. In this technical report, we focus on visual representations. We review and evaluate self-supervised methods to leverage unlabeled images and pretrain a model, which we then fine-tune on a custom VQA task that allows controlled evaluation and diagnosis. We compare energy-based models (EBMs) with contrastive learning (CL). While EBMs are growing in popularity, they lack an evaluation on downstream tasks. We find that both EBMs and CL can learn representations from unlabeled images that enable training a VQA model on very little annotated data. In a simple setting similar to CLEVR, we find that CL representations also improve systematic generalization, and even match the performance of representations from a larger, supervised, ImageNet-pretrained model. However, we find EBMs to be difficult to train because of instabilities and high variability in their results. Although EBMs prove useful for OOD detection, other results on supervised energy-based training and uncertainty calibration are largely negative. Overall, CL currently seems a preferable option over EBMs.
Geometry parameter estimation for sparse X-ray log imaging
Authors: Angelina Senchukova, Jarkko Suuronen, Jere Heikkinen, Lassi Roininen
Subjects: Computational Engineering, Finance, and Science (cs.CE); Image and Video Processing (eess.IV)
Abstract
We consider geometry parameter estimation in industrial sawmill fan-beam X-ray tomography. In such industrial settings, scanners do not always allow identification of the location of the source-detector pair, which creates the issue of unknown geometry. This work considers two approaches for geometry estimation. Our first approach is a calibration object correlation method in which we calculate the maximum cross-correlation between a known-sized calibration object image and its filtered backprojection reconstruction and use differential evolution as an optimiser. The second approach is projection trajectory simulation, where we use a set of known intersection points and a sequential Monte Carlo method for estimating the posterior density of the parameters. We show numerically that a large set of parameters can be used for artefact-free reconstruction. We deploy Bayesian inversion with Cauchy priors for synthetic and real sawmill data for detection of knots with a very low number of measurements and uncertain measurement geometry.
Uncertainty-aware Panoptic Segmentation
Authors: Kshitij Sirohi, Sajad Marvi, Daniel Büscher, Wolfram Burgard
Subjects: Computer Vision and Pattern Recognition (cs.CV); Robotics (cs.RO)
Abstract
Reliable scene understanding is indispensable for modern autonomous systems. Current learning-based methods typically try to maximize their performance based on segmentation metrics that only consider the quality of the segmentation. However, for the safe operation of a system in the real world it is crucial to consider the uncertainty in the prediction as well. In this work, we introduce the novel task of uncertainty-aware panoptic segmentation, which aims to predict per-pixel semantic and instance segmentations, together with per-pixel uncertainty estimates. We define two novel metrics to facilitate its quantitative analysis, the uncertainty-aware Panoptic Quality (uPQ) and the panoptic Expected Calibration Error (pECE). We further propose the novel top-down Evidential Panoptic Segmentation Network (EvPSNet) to solve this task. Our architecture employs a simple yet effective probabilistic fusion module that leverages the predicted uncertainties. Additionally, we propose a new Lov\'asz evidential loss function to optimize the IoU for the segmentation utilizing the probabilities provided by deep evidential learning. Furthermore, we provide several strong baselines combining state-of-the-art panoptic segmentation networks with sampling-free uncertainty estimation techniques. Extensive evaluations show that our EvPSNet achieves the new state-of-the-art for the standard Panoptic Quality (PQ), as well as for our uncertainty-aware panoptic metrics.
Keyword: out of distribution detection
There is no result
Keyword: out-of-distribution detection
RegMixup: Mixup as a Regularizer Can Surprisingly Improve Accuracy and Out Distribution Robustness
Keyword: expected calibration error
Uncertainty-aware Panoptic Segmentation
Keyword: overconfident
There is no result
Keyword: overconfidence
There is no result
Keyword: confidence
Semi-supervised Contrastive Outlier removal for Pseudo Expectation Maximization (SCOPE)
Collecting high-quality adversarial data for machine reading comprehension tasks with humans and models in the loop
What Can Secondary Predictions Tell Us? An Exploration on Question-Answering with SQuAD-v2.0
Constructing MDP Abstractions Using Data with Formal Guarantees
Mining Seasonal Temporal Patterns in Big Time Series
Two-Stage Neural Contextual Bandits for Personalised News Recommendation
Keyword: scaling
NumS: Scalable Array Programming for the Cloud
Left Heavy Tails and the Effectiveness of the Policy and Value Networks in DNN-based best-first search for Sokoban Planning
Beyond neural scaling laws: beating power law scaling via data pruning
Keyword: calibration
EBMs vs. CL: Exploring Self-Supervised Visual Pretraining for Visual Question Answering
Geometry parameter estimation for sparse X-ray log imaging
Uncertainty-aware Panoptic Segmentation