Classification performance is often not uniform over the data. Some areas in the input space are easier to classify than others. Features that hold information about the "difficulty" of the data may be non-discriminative and are therefore disregarded in the classification process. We propose a meta-learning approach where performance may be improved by post-processing. This improvement is done by establishing a dynamic threshold on the base-classifier results. Since the base-classifier is treated as a "black box" the method presented can be used on any state of the art classifier in order to try an improve its performance. We focus our attention on how to better control the true-positive/false-positive trade-off known as the ROC curve. We propose an algorithm for the derivation of optimal thresholds by redistributing the error depending on features that hold information about difficulty. We demonstrate the resulting benefit on both synthetic and real-life data.
This paper presents a novel approach to visual objects classification based on generating simple fuzzy classifiers using local image features to distinguish between one known class and other classes. Boosting meta learning is used to find the most representative local features. The proposed approach is tested on a state-of-the-art image dataset and compared with the bag-of-features image representation model combined with the Support Vector Machine classification. The novel method gives better classification accuracy and the time of learning and testing process is more than 30% shorter.
Online model-free reinforcement learning (RL) methods with continuous actions are playing a prominent role when dealing with real-world applications such as Robotics. However, when confronted to non-stationary environments, these methods crucially rely on an exploration-exploitation trade-off which is rarely dynamically and automatically adjusted to changes in the environment. Here we propose an active exploration algorithm for RL in structured (parameterized) continuous action space. This framework deals with a set of discrete actions, each of which is parameterized with continuous variables. Discrete exploration is controlled through a Boltzmann softmax function with an inverse temperature $\beta$ parameter. In parallel, a Gaussian exploration is applied to the continuous action parameters. We apply a meta-learning algorithm based on the comparison between variations of short-term and long-term reward running averages to simultaneously tune $\beta$ and the width of the Gaussian distribution from which continuous action parameters are drawn. When applied to a simple virtual human-robot interaction task, we show that this algorithm outperforms continuous parameterized RL both without active exploration and with active exploration based on uncertainty variations measured by a Kalman-Q-learning algorithm.
Meta-learning consists in learning learning algorithms. We use a Long Short Term Memory (LSTM) based network to learn to compute on-line updates of the parameters of another neural network. These parameters are stored in the cell state of the LSTM. Our framework allows to compare learned algorithms to hand-made algorithms within the traditional train and test methodology. In an experiment, we learn a learning algorithm for a one-hidden layer Multi-Layer Perceptron (MLP) on non-linearly separable datasets. The learned algorithm is able to update parameters of both layers and generalise well on similar datasets.
One of the main obstacles to broad application of reinforcement learning methods is the parameter sensitivity of our core learning algorithms. In many large-scale applications, online computation and function approximation represent key strategies in scaling up reinforcement learning algorithms. In this setting, we have effective and reasonably well understood algorithms for adapting the learning-rate parameter, online during learning. Such meta-learning approaches can improve robustness of learning and enable specialization to current task, improving learning speed. For temporal-difference learning algorithms which we study here, there is yet another parameter, $\lambda$, that similarly impacts learning speed and stability in practice. Unfortunately, unlike the learning-rate parameter, $\lambda$ parametrizes the objective function that temporal-difference methods optimize. Different choices of $\lambda$ produce different fixed-point solutions, and thus adapting $\lambda$ online and characterizing the optimization is substantially more complex than adapting the learning-rate parameter. There are no meta-learning method for $\lambda$ that can achieve (1) incremental updating, (2) compatibility with function approximation, and (3) maintain stability of learning under both on and off-policy sampling. In this paper we contribute a novel objective function for optimizing $\lambda$ as a function of state rather than time. We derive a new incremental, linear complexity $\lambda$-adaption algorithm that does not require offline batch updating or access to a model of the world, and present a suite of experiments illustrating the practicality of our new algorithm in three different settings. Taken together, our contributions represent a concrete step towards black-box application of temporal-difference learning methods in real world problems.
Extracting actionable intelligence from distributed, heterogeneous, correlated and high-dimensional data sources requires run-time processing and learning both locally and globally. In the last decade, a large number of meta-learning techniques have been proposed in which local learners make online predictions based on their locally-collected data instances, and feed these predictions to an ensemble learner, which fuses them and issues a global prediction. However, most of these works do not provide performance guarantees or, when they do, these guarantees are asymptotic. None of these existing works provide confidence estimates about the issued predictions or rate of learning guarantees for the ensemble learner. In this paper, we provide a systematic ensemble learning method called Hedged Bandits, which comes with both long run (asymptotic) and short run (rate of learning) performance guarantees. Moreover, our approach yields performance guarantees with respect to the optimal local prediction strategy, and is also able to adapt its predictions in a data-driven manner. We illustrate the performance of Hedged Bandits in the context of medical informatics and show that it outperforms numerous online and offline ensemble learning methods.
The move from hand-designed features to learned features in machine learning has been wildly successful. In spite of this, optimization algorithms are still designed by hand. In this paper we show how the design of an optimization algorithm can be cast as a learning problem, allowing the algorithm to learn to exploit structure in the problems of interest in an automatic way. Our learned algorithms, implemented by LSTMs, outperform generic, hand-designed competitors on the tasks for which they are trained, and also generalize well to new tasks with similar structure. We demonstrate this on a number of tasks, including simple convex problems, training neural networks, and styling images with neural art.
The rapid advancement of machine learning techniques has re-energized research into general artificial intelligence. While the idea of domain-agnostic meta-learning is appealing, this emerging field must come to terms with its relationship to human cognition and the statistics and structure of the tasks humans perform. The position of this article is that only by aligning our agents' abilities and environments with those of humans do we stand a chance at developing general artificial intelligence (GAI). A broad reading of the famous 'No Free Lunch' theorem is that there is no universally optimal inductive bias or, equivalently, bias-free learning is impossible. This follows from the fact that there are an infinite number of ways to extrapolate data, any of which might be the one used by the data generating environment; an inductive bias prefers some of these extrapolations to others, which lowers performance in environments using these adversarial extrapolations. We may posit that the optimal GAI is the one that maximally exploits the statistics of its environment to create its inductive bias; accepting the fact that this agent is guaranteed to be extremely sub-optimal for some alternative environments. This trade-off appears benign when thinking about the environment as being the physical universe, as performance on any fictive universe is obviously irrelevant. But, we should expect a sharper inductive bias if we further constrain our environment. Indeed, we implicitly do so by defining GAI in terms of accomplishing that humans consider useful. One common version of this is need the for 'common-sense reasoning', which implicitly appeals to the statistics of physical universe as perceived by humans.
In recent years deep reinforcement learning (RL) systems have attained superhuman performance in a number of challenging task domains. However, a major limitation of such applications is their demand for massive amounts of training data. A critical present objective is thus to develop deep RL methods that can adapt rapidly to new tasks. In the present work we introduce a novel approach to this challenge, which we refer to as deep meta-reinforcement learning. Previous work has shown that recurrent networks can support meta-learning in a fully supervised context. We extend this approach to the RL setting. What emerges is a system that is trained using one RL algorithm, but whose recurrent dynamics implement a second, quite separate RL procedure. This second, learned RL algorithm can differ from the original one in arbitrary ways. Importantly, because it is learned, it is configured to exploit structure in the training domain. We unpack these points in a series of seven proof-of-concept experiments, each of which examines a key aspect of deep meta-RL. We consider prospects for extending and scaling up the approach, and also point out some potentially important implications for neuroscience.
This paper reports on a writing style analysis of hyperpartisan (i.e., extremely one-sided) news in connection to fake news. It presents a large corpus of 1,627 articles that were manually fact-checked by professional journalists from BuzzFeed. The articles originated from 9 well-known political publishers, 3 each from the mainstream, the hyperpartisan left-wing, and the hyperpartisan right-wing. In sum, the corpus contains 299 fake news, 97% of which originated from hyperpartisan publishers. We propose and demonstrate a new way of assessing style similarity between text categories via Unmasking---a meta-learning approach originally devised for authorship verification---, revealing that the style of left-wing and right-wing news have a lot more in common than any of the two have with the mainstream. Furthermore, we show that hyperpartisan news can be discriminated well by its style from the mainstream (F1=0.78), as can be satire from both (F1=0.81). Unsurprisingly, style-based fake news detection does not live up to scratch (F1=0.46). Nevertheless, the former results are important to implement pre-screening for fake news detectors.
The efficiency of reinforcement learning algorithms depends critically on a few meta-parameters that modulates the learning updates and the trade-off between exploration and exploitation. The adaptation of the meta-parameters is an open question in reinforcement learning, which arguably has become more of an issue recently with the success of deep reinforcement learning in high-dimensional state spaces. The long learning times in domains such as Atari 2600 video games makes it not feasible to perform comprehensive searches of appropriate meta-parameter values. We propose the Online Meta-learning by Parallel Algorithm Competition (OMPAC) method. In the OMPAC method, several instances of a reinforcement learning algorithm are run in parallel with small differences in the initial values of the meta-parameters. After a fixed number of episodes, the instances are selected based on their performance in the task at hand. Before continuing the learning, Gaussian noise is added to the meta-parameters with a predefined probability. We validate the OMPAC method by improving the state-of-the-art results in stochastic SZ-Tetris and in standard Tetris with a smaller, 10$\times$10, board, by 31% and 84%, respectively, and by improving the results for deep Sarsa($\lambda$) agents in three Atari 2600 games by 62% or more. The experiments also show the ability of the OMPAC method to adapt the meta-parameters according to the learning progress in different tasks.
This extended abstract describes the participation of RECOD Titans in parts 1 and 3 of the ISIC Challenge 2017 "Skin Lesion Analysis Towards Melanoma Detection" (ISBI 2017). Although our team has a long experience with melanoma classification, the ISIC Challenge 2017 was the very first time we worked on skin-lesion segmentation. For part 1 (segmentation), our final submission used four of our models: two trained with all 2000 samples, without a validation split, for 250 and for 500 epochs respectively; and other two trained and validated with two different 1600/400 splits, for 220 epochs. Those four models, individually, achieved between 0.780 and 0.783 official validation scores. Our final submission averaged the output of those four models achieved a score of 0.793. For part 3 (classification), the submitted test run as well as our last official validation run were the result from a meta-model that assembled seven base deep-learning models: three based on Inception-V4 trained on our largest dataset; three based on Inception trained on our smallest dataset; and one based on ResNet-101 trained on our smaller dataset. The results of those component models were stacked in a meta-learning layer based on an SVM trained on the validation set of our largest dataset.
Outlier detection is the identification of points in a dataset that do not conform to the norm. Outlier detection is highly sensitive to the choice of the detection algorithm and the feature subspace used by the algorithm. Extracting domain-relevant insights from outliers needs systematic exploration of these choices since diverse outlier sets could lead to complementary insights. This challenge is especially acute in an interactive setting, where the choices must be explored in a time-constrained manner. In this work, we present REMIX, the first system to address the problem of outlier detection in an interactive setting. REMIX uses a novel mixed integer programming (MIP) formulation for automatically selecting and executing a diverse set of outlier detectors within a time limit. This formulation incorporates multiple aspects such as (i) an upper limit on the total execution time of detectors (ii) diversity in the space of algorithms and features, and (iii) meta-learning for evaluating the cost and utility of detectors. REMIX provides two distinct ways for the analyst to consume its results: (i) a partitioning of the detectors explored by REMIX into perspectives through low-rank non-negative matrix factorization; each perspective can be easily visualized as an intuitive heatmap of experiments versus outliers, and (ii) an ensembled set of outliers which combines outlier scores from all detectors. We demonstrate the benefits of REMIX through extensive empirical validation on real-world data.
One of the long-standing challenges in Artificial Intelligence for learning goal-directed behavior is to build a single agent which can solve multiple tasks. Recent progress in multi-task learning for goal-directed sequential problems has been in the form of distillation based learning wherein a student network learns from multiple task-specific expert networks by mimicking the task-specific policies of the expert networks. While such approaches offer a promising solution to the multi-task learning problem, they require supervision from large expert networks which require extensive data and computation time for training. In this work, we propose an efficient multi-task learning framework which solves multiple goal-directed tasks in an on-line setup without the need for expert supervision. Our work uses active learning principles to achieve multi-task learning by sampling the harder tasks more than the easier ones. We propose three distinct models under our active sampling framework. An adaptive method with extremely competitive multi-tasking performance. A UCB-based meta-learner which casts the problem of picking the next task to train on as a multi-armed bandit problem. A meta-learning method that casts the next-task picking problem as a full Reinforcement Learning problem and uses actor critic methods for optimizing the multi-tasking performance directly. We demonstrate results in the Atari 2600 domain on seven multi-tasking instances: three 6-task instances, one 8-task instance, two 12-task instances and one 21-task instance.
Neural networks have been successfully applied in applications with a large amount of labeled data. However, the task of rapid generalization on new concepts with small training data while preserving performances on previously learned ones still presents a significant challenge to neural network models. In this work, we introduce a novel meta learning method, Meta Networks (MetaNet), that learns a meta-level knowledge across tasks and shifts its inductive biases via fast parameterization for rapid generalization. When evaluated on Omniglot and Mini-ImageNet benchmarks, our MetaNet models achieve a near human-level performance and outperform the baseline approaches by up to 6% accuracy. We demonstrate several appealing properties of MetaNet relating to generalization and continual learning.
The success of automated driving deployment is highly depending on the ability to develop an efficient and safe driving policy. The problem is well formulated under the framework of optimal control as a cost optimization problem. Model based solutions using traditional planning are efficient, but require the knowledge of the environment model. On the other hand, model free solutions suffer sample inefficiency and require too many interactions with the environment, which is infeasible in practice. Methods under the Reinforcement Learning framework usually require the notion of a reward function, which is not available in the real world. Imitation learning helps in improving sample efficiency by introducing prior knowledge obtained from the demonstrated behavior, on the risk of exact behavior cloning without generalizing to unseen environments. In this paper we propose a Meta learning framework, based on data set aggregation, to improve generalization of imitation learning algorithms. Under the proposed framework, we propose MetaDAgger, a novel algorithm which tackles the generalization issues in traditional imitation learning. We use The Open Race Car Simulator (TORCS) to test our algorithm. Results on unseen test tracks show significant improvement over traditional imitation learning algorithms, improving the learning time and sample efficiency in the same time. The results are also supported by visualization of the learnt features to prove generalization of the captured details.
Supervised learning is ubiquitous in medical image analysis. In this paper we consider the problem of meta-learning -- predicting which methods will perform well in an unseen classification problem, given previous experience with other classification problems. We investigate the first step of such an approach: how to quantify the similarity of different classification problems. We characterize datasets sampled from six classification problems by performance ranks of simple classifiers, and define the similarity by the inverse of Euclidean distance in this meta-feature space. We visualize the similarities in a 2D space, where meaningful clusters start to emerge, and show that the proposed representation can be used to classify datasets according to their origin with 89.3\% accuracy. These findings, together with the observations of recent trends in machine learning, suggest that meta-learning could be a valuable tool for the medical imaging community.
We propose prototypical networks for the problem of few-shot classification, where a classifier must generalize to new classes not seen in the training set, given only a small number of examples of each new class. Prototypical networks learn a metric space in which classification can be performed by computing distances to prototype representations of each class. Compared to recent approaches for few-shot learning, they reflect a simpler inductive bias that is beneficial in this limited-data regime, and achieve excellent results. We provide an analysis showing that some simple design decisions can yield substantial improvements over recent approaches involving complicated architectural choices and meta-learning. We further extend prototypical networks to zero-shot learning and achieve state-of-the-art results on the CU-Birds dataset.
We propose a novel and flexible approach to meta-learning for learning-to-learn from only a few examples. Our framework is motivated by actor-critic reinforcement learning, but can be applied to both reinforcement and supervised learning. The key idea is to learn a meta-critic: an action-value function neural network that learns to criticise any actor trying to solve any specified task. For supervised learning, this corresponds to the novel idea of a trainable task-parametrised loss generator. This meta-critic approach provides a route to knowledge transfer that can flexibly deal with few-shot and semi-supervised conditions for both reinforcement and supervised learning. Promising results are shown on both reinforcement and supervised learning problems.
We consider the problem of learning when obtaining the training labels is costly, which is usually tackled in the literature using active-learning techniques. These approaches provide strategies to choose the examples to label before or during training. These strategies are usually based on heuristics or even theoretical measures, but are not learned as they are directly used during training. We design a model which aims at \textit{learning active-learning strategies} using a meta-learning setting. More specifically, we consider a pool-based setting, where the system observes all the examples of the dataset of a problem and has to choose the subset of examples to label in a single shot. Experiments show encouraging results.