We introduce Mintaka, a complex, natural, and multilingual dataset designed for experimenting with end-to-end question-answering models. Mintaka is composed of 20,000 question-answer pairs collected in English, annotated with Wikidata entities, and translated into Arabic, French, German, Hindi, Italian, Japanese, Portuguese, and Spanish for a total of 180,000 samples. Mintaka includes 8 types of complex questions, including superlative, intersection, and multi-hop questions, which were naturally elicited from crowd workers. We run baselines over Mintaka, the best of which achieves 38% hits@1 in English and 31% hits@1 multilingually, showing that existing models have room for improvement. We release Mintaka at https://github.com/amazon-research/mintaka.

Recently, end-to-end (E2E) trained models for question answering over knowledge graphs (KGQA) have delivered promising results using only a weakly supervised dataset.However, these models are trained and evaluated in a setting where hand-annotated question entities are supplied to the model, leaving the important and non-trivial task of entity resolution (ER) outside the scope of E2E learning.In this work, we extend the boundaries of E2E learning for KGQA to include the training of an ER component.Our model only needs the question text and the answer entities to train, and delivers a stand-alone QA model that does not require an additional ER component to be supplied during runtime.Our approach is fully differentiable, thanks to its reliance on a recent method for building differentiable KGs (Cohen et al., 2020).We evaluate our E2E trained model on two public datasets and show that it comes close to baseline models that use handannotated entities.

End-to-end question answering using a differentiable knowledge graph is a promising technique that requires only weak supervision, produces interpretable results, and is fully differentiable.Previous implementations of this technique (Cohen et al., 2020) have focused on single-entity questions using a relation following operation.In this paper, we propose a model that explicitly handles multiple-entity questions by implementing a new intersection operation, which identifies the shared elements between two sets of entities.We find that introducing intersection improves performance over a baseline model on two datasets, WebQuestionsSP (69.6% to 73.3% Hits@1) and ComplexWebQuestions (39.8% to 48.7% Hits@1), and in particular, improves performance on questions with multiple entities by over 14% on WebQuestionsSP and by 19% on ComplexWebQuestions.

End-to-end neural data-to-text (D2T) generation has recently emerged as an alternative to pipeline-based architectures.However, it has faced challenges generalizing to new domains and generating semantically consistent text.In this work, we present DATATUNER, a neural, end-to-end data-to-text generation system that makes minimal assumptions about the data representation and target domain.We take a two-stage generation-reranking approach, combining a fine-tuned language model with a semantic fidelity classifier.Each component is learnt end-toend without needing dataset-specific heuristics, entity delexicalization, or post-processing.We show that DATATUNER achieves state of the art results on automated metrics across four major D2T datasets (LDC2017T10, WebNLG, ViGGO, and Cleaned E2E), with fluency assessed by human annotators as nearing or exceeding the human-written reference texts.Our generated text has better semantic fidelity than the state of the art on these datasets.We further demonstrate that our model-based semantic fidelity scorer is a better assessment tool compared to traditional heuristic-based measures of semantic accuracy.

While models have reached superhuman performance on popular question answering (QA) datasets such as SQuAD, they have yet to outperform humans on the task of question answering itself.In this paper, we investigate if models are learning reading comprehension from QA datasets by evaluating BERT-based models across five datasets.We evaluate models on their generalizability to out-of-domain examples, responses to missing or incorrect data, and ability to handle question variations.We find that no single dataset is robust to all of our experiments and identify shortcomings in both datasets and evaluation methods.Following our analysis, we make recommendations for building future QA datasets that better evaluate the task of question answering through reading comprehension.We also release code to convert QA datasets to a shared format for easier experimentation at https: //github.com/amazon-research/ qa-dataset-converter.

Adaptive tracking-by-detection methods are widely used in computer vision for tracking arbitrary objects. Current approaches treat the tracking problem as a classification task and use online learning techniques to update the object model. However, for these updates to happen one needs to convert the estimated object position into a set of labelled training examples, and it is not clear how best to perform this intermediate step. Furthermore, the objective for the classifier (label prediction) is not explicitly coupled to the objective for the tracker (estimation of object position). In this paper, we present a framework for adaptive visual object tracking based on structured output prediction. By explicitly allowing the output space to express the needs of the tracker, we avoid the need for an intermediate classification step. Our method uses a kernelised structured output support vector machine (SVM), which is learned online to provide adaptive tracking. To allow our tracker to run at high frame rates, we (a) introduce a budgeting mechanism that prevents the unbounded growth in the number of support vectors that would otherwise occur during tracking, and (b) show how to implement tracking on the GPU. Experimentally, we show that our algorithm is able to outperform state-of-the-art trackers on various benchmark videos. Additionally, we show that we can easily incorporate additional features and kernels into our framework, which results in increased tracking performance.

This paper introduces a novel classification method termed Alternating Decision Forests (ADFs), which formulates the training of Random Forests explicitly as a global loss minimization problem. During training, the losses are minimized via keeping an adaptive weight distribution over the training samples, similar to Boosting methods. In order to keep the method as flexible and general as possible, we adopt the principle of employing gradient descent in function space, which allows to minimize arbitrary losses. Contrary to Boosted Trees, in our method the loss minimization is an inherent part of the tree growing process, thus allowing to keep the benefits of common Random Forests, such as, parallel processing. We derive the new classifier and give a discussion and evaluation on standard machine learning data sets. Furthermore, we show how ADFs can be easily integrated into an object detection application. Compared to both, standard Random Forests and Boosted Trees, ADFs give better performance in our experiments, while yielding more compact models in terms of tree depth.

Efficient keypoint-based object detection methods are used in many real-time computer vision applications. These approaches often model an object as a collection of keypoints and associated descriptors, and detection then involves first constructing a set of correspondences between object and image keypoints via descriptor matching, and subsequently using these correspondences as input to a robust geometric estimation algorithm such as RANSAC to find the transformation of the object in the image. In such approaches, the object model is generally constructed offline, and does not adapt to a given environment at runtime. Furthermore, the feature matching and transformation estimation stages are treated entirely separately. In this paper, we introduce a new approach to address these problems by combining the overall pipeline of correspondence generation and transformation estimation into a single structured output learning framework. Following the recent trend of using efficient binary descriptors for feature matching, we also introduce an approach to approximate the learned object model as a collection of binary basis functions which can be evaluated very efficiently at runtime. Experiments on challenging video sequences show that our algorithm significantly improves over state-of-the-art descriptor matching techniques using a range of descriptors, as well as recent online learning based approaches.

Current state-of-the-art object classification systems are trained using large amounts of hand-labeled images. In this paper, we present an approach that shows how to use unlabeled video sequences, comprising weakly-related object categories towards the target class, to learn better classifiers for tracking and detection. The underlying idea is to exploit the space-time consistency of moving objects to learn classifiers that are robust to local transformations. In particular, we use dense optical flow to find moving objects in videos in order to train part-based random forests that are insensitive to natural transformations. Our method, which is called Video Forests, can be used in two settings: first, labeled training data can be regularized to force the trained classifier to generalize better towards small local transformations. Second, as part of a tracking-by-detection approach, it can be used to train a general codebook solely on pair-wise data that can then be applied to tracking of instances of a priori unknown object categories. In the experimental part, we show on benchmark datasets for both tracking and detection that incorporating unlabeled videos into the learning of visual classifiers leads to improved results.

Adaptive tracking-by-detection methods are widely used in computer vision for tracking arbitrary objects. Current approaches treat the tracking problem as a classification task and use online learning techniques to update the object model. However, for these updates to happen one needs to convert the estimated object position into a set of labelled training examples, and it is not clear how best to perform this intermediate step. Furthermore, the objective for the classifier (label prediction) is not explicitly coupled to the objective for the tracker (accurate estimation of object position). In this paper, we present a framework for adaptive visual object tracking based on structured output prediction. By explicitly allowing the output space to express the needs of the tracker, we are able to avoid the need for an intermediate classification step. Our method uses a kernelized structured output support vector machine (SVM), which is learned online to provide adaptive tracking. To allow for real-time application, we introduce a budgeting mechanism which prevents the unbounded growth in the number of support vectors which would otherwise occur during tracking. Experimentally, we show that our algorithm is able to outperform state-of-the-art trackers on various benchmark videos. Additionally, we show that we can easily incorporate additional features and kernels into our framework, which results in increased performance.

Local Coordinate Coding (LCC) [18] is a method for modeling functions of data lying on non-linear manifolds. It provides a set of anchor points which form a local coordinate system, such that each data point on the manifold can be approximated by a linear combination of its anchor points, and the linear weights become the local coordinate coding. In this paper we propose encoding data using orthogonal anchor planes, rather than anchor points. Our method needs only a few orthogonal anchor planes for coding, and it can linearize any (\alpha,\beta,p)-Lipschitz smooth nonlinear function with a fixed expected value of the upper-bound approximation error on any high dimensional data. In practice, the orthogonal coordinate system can be easily learned by minimizing this upper bound using singular value decomposition (SVD). We apply our method to model the coordinates locally in linear SVMs for classification tasks, and our experiment on MNIST shows that using only 50 anchor planes our method achieves 1.72% error rate, while LCC achieves 1.90% error rate using 4096 anchor points.

Online boosting is one of the most successful online learning algorithms in computer vision. While many challenging online learning problems are inherently multi-class, online boosting and its variants are only able to solve binary tasks. In this paper, we present Online Multi-Class LPBoost (OMCLP) which is directly applicable to multi-class problems. From a theoretical point of view, our algorithm tries to maximize the multi-class soft-margin of the samples. In order to solve the LP problem in online settings, we perform an efficient variant of online convex programming, which is based on primal-dual gradient descent-ascent update strategies. We conduct an extensive set of experiments over machine learning benchmark datasets, as well as, on Caltech 101 category recognition dataset. We show that our method is able to outperform other online multi-class methods. We also apply our method to tracking where, we present an intuitive way to convert the binary tracking by detection problem to a multi-class problem where background patterns which are similar to the target class, become virtual classes. Applying our novel model, we outperform or achieve the state-of-the-art results on benchmark tracking videos.

Tracking-by-detection is increasingly popular in order to tackle the visual tracking problem. Existing adaptive methods suffer from the drifting problem, since they rely on self-updates of an on-line learning method. In contrast to previous work that tackled this problem by employing semi-supervised or multiple-instance learning, we show that augmenting an on-line learning method with complementary tracking approaches can lead to more stable results. In particular, we use a simple template model as a non-adaptive and thus stable component, a novel optical-flow-based mean-shift tracker as highly adaptive element and an on-line random forest as moderately adaptive appearance-based learner. We combine these three trackers in a cascade. All of our components run on GPUs or similar multi-core systems, which allows for real-time performance. We show the superiority of our system over current state-of-the-art tracking methods in several experiments on publicly available data.

A recent dominating trend in tracking called tracking-by-detection uses on-line classifiers in order to redetect objects over succeeding frames. Although these methods usually deliver excellent results and run in real-time they also tend to drift in case of wrong updates during the self-learning process. Recent approaches tackled this problem by formulating tracking-by-detection as either one-shot semi-supervised learning or multiple instance learning. Semi-supervised learning allows for incorporating priors and is more robust in case of occlusions while multiple-instance learning resolves the uncertainties where to take positive updates during tracking. In this work, we propose an on-line semi-supervised learning algorithm which is able to combine both of these approaches into a coherent framework. This leads to more robust results than applying both approaches separately. Additionally, we introduce a combined loss that simultaneously uses labeled and unlabeled samples, which makes our tracker more adaptive compared to previous on-line semi-supervised methods. Experimentally, we demonstrate that by using our semi-supervised multiple-instance approach and utilizing robust learning methods, we are able to outperform state-of-the-art methods on various benchmark tracking videos.

Randomized learning methods (i.e., Forests or Ferns) have shown excellent capabilities for various computer vision applications. However, it was shown that the tree structure in Forests can be replaced by even simpler structures, e.g., Random Naive Bayes classifiers, yielding similar performance. The goal of this paper is to benefit from these findings to develop an efficient on-line learner. Based on the principals of on-line Random Forests, we adapt the Random Naive Bayes classifier to the on-line domain. For that purpose, we propose to use on-line histograms as weak learners, which yield much better performance than simple decision stumps. Experimentally we show, that the approach is applicable to incremental learning on machine learning datasets. Additionally, we propose to use an IIR filtering-like forgetting function for the weak learners to enable adaptivity and evaluate our classifier on the task of tracking by detection.

Common methods for interactive texture segmentation rely on probability maps based on low dimensional features such as e.g. intensity or color, that are usually modeled using basic learning algorithms such as histograms or Gaussian Mixture Models. The use of low level features allows for fast generation of these hypotheses but limits applicability to a small class of images. We address this problem by learning complex descriptors with Random Forests and exploiting their inherent parallelism in a GPU implementation. The segmentation itself is based on a convex energy functional that uses weighted Total Variation regularization and a point-wise data term allowing for continuous foreground/background membership hypotheses. Its globally optimal solution is obtained by a fast primal-dual algorithm providing a reasonable convergence criterion. As a result, we present a versatile interactive texture segmentation framework. We show experiments with natural, artificial and medical data and demonstrate superior results compared to two recent approaches.

Many semi-supervised learning algorithms only deal with binary classification. Their extension to the multi-class problem is usually obtained by repeatedly solving a set of binary problems. Additionally, many of these methods do not scale very well with respect to a large number of unlabeled samples, which limits their applications to large-scale problems with many classes and unlabeled samples. In this paper, we directly address the multi-class semi-supervised learning problem by an efficient boosting method. In particular, we introduce a new multi-class margin-maximizing loss function for the unlabeled data and use the generalized expectation regularization for incorporating cluster priors into the model. Our approach enables efficient usage of very large data sets. The performance and efficiency of our method is demonstrated on both standard machine learning data sets as well as on challenging object categorization tasks.

Random Forests (RFs) have become commonplace in many computer vision applications. Their popularity is mainly driven by their high computational efficiency during both training and evaluation while still being able to achieve state-of-the-art accuracy. This work extends the usage of Random Forests to Semi-Supervised Learning (SSL) problems. We show that traditional decision trees are optimizing multi-class margin maximizing loss functions. From this intuition, we develop a novel multi-class margin definition for the unlabeled data, and an iterative deterministic annealing-style training algorithm maximizing both the multi-class margin of labeled and unlabeled samples. In particular, this allows us to use the predicted labels of the unlabeled data as additional optimization variables. Furthermore, we propose a control mechanism based on the out-of-bag error, which prevents the algorithm from degradation if the unlabeled data is not useful for the task. Our experiments demonstrate state-of-the-art semi-supervised learning performance in typical machine learning problems and constant improvement using unlabeled data for the Caltech-101 object categorization task.

"When everything fails, ask for additional domain knowledge" is the current motto of machine learning. Therefore, assessing the real added value of prior/domain knowledge is a both deep and practical question. Most commercial data mining programs accept data pre-formatted as a table, each example being encoded as a fixed set of features. Is it worth spending time engineering elaborate features incorporating domain knowledge and/or designing ad hoc algorithms? Or else, can off-the-shelf programs working on simple features encoding the raw data without much domain knowledge do as well or better than skilled data analysts? To answer these questions, we organized a challenge for IJCNN 2007. The participants were allowed to compete in two tracks: The "prior knowledge" (PK) track, for which they had access to the original raw data representation and as much knowledge as possible about the data, and the "agnostic learning" (AL) track for which they were forced to use data pre-formatted as a table with dummy features. The AL vs. PK challenge Web site remains open: http://www.agnostic.inf.ethz.ch/.

A major challenge for machine learning algorithms in real world applications is to predict their performance. We have approached this question by organizing a challenge in performance prediction for WCCI 2006. The class of problems addressed are classification problems encountered in pattern recognition (classification of images, speech recognition), medical diagnosis, marketing (customer categorization), text categorization (filtering of spam). Over 100 participants have been trying to build the best possible classifier from training data and guess their generalization error on a large unlabeled test set. The challenge scores indicate that cross-validation yields good results both for model selection and performance prediction. Alternative model selection strategies were also sometimes employed with success. The challenge web site keeps open for post-challenge submissions: http://www.modelselect.inf.ethz.ch/.