A core issue with learning to optimize neural networks has been the lack of generalization to real world problems. To address this, we describe a system designed from a generalization-first perspective, learning to update optimizer hyperparameters instead of model parameters directly using novel features, actions, and a reward function. This system outperforms Adam at all neural network tasks including on modalities not seen during training. We achieve 2x speedups on ImageNet, and a 2.5x speedup on a language modeling task using over 5 orders of magnitude more compute than the training tasks.

One-pixel attack is a curious way of deceiving neural network classifier by changing only one pixel in the input image. The full potential and boundaries of this attack method are not yet fully understood. In this research, the successful and unsuccessful attacks are studied in more detail to illustrate the working mechanisms of a one-pixel attack created using differential evolution. The data comes from our earlier studies where we applied the attack against medical imaging. We used a real breast cancer tissue dataset and a real classifier as the attack target. This research presents ways to analyze chromatic and spatial distributions of one-pixel attacks. In addition, we present one-pixel attack confidence maps to illustrate the behavior of the target classifier. We show that the more effective attacks change the color of the pixel more, and that the successful attacks are situated at the center of the images. This kind of analysis is not only useful for understanding the behavior of the attack but also the qualities of the classifying neural network.

When faced with learning challenging new tasks, humans often follow sequences of steps that allow them to incrementally build up the necessary skills for performing these new tasks. However, in machine learning, models are most often trained to solve the target tasks directly.Inspired by human learning, we propose a novel curriculum learning approach which decomposes challenging tasks into sequences of easier intermediate goals that are used to pre-train a model before tackling the target task. We focus on classification tasks, and design the intermediate tasks using an automatically constructed label hierarchy. We train the model at each level of the hierarchy, from coarse labels to fine labels, transferring acquired knowledge across these levels. For instance, the model will first learn to distinguish animals from objects, and then use this acquired knowledge when learning to classify among more fine-grained classes such as cat, dog, car, and truck. Most existing curriculum learning algorithms for supervised learning consist of scheduling the order in which the training examples are presented to the model. In contrast, our approach focuses on the output space of the model. We evaluate our method on several established datasets and show significant performance gains especially on classification problems with many labels. We also evaluate on a new synthetic dataset which allows us to study multiple aspects of our method.

Inverse molecular design, i.e., designing molecules with specific target properties, can be posed as an optimization problem. High-dimensional optimization tasks in the natural sciences are commonly tackled via population-based metaheuristic optimization algorithms such as evolutionary algorithms. However, expensive property evaluation, which is often required, can limit the widespread use of such approaches as the associated cost can become prohibitive. Herein, we present JANUS, a genetic algorithm that is inspired by parallel tempering. It propagates two populations, one for exploration and another for exploitation, improving optimization by reducing expensive property evaluations. Additionally, JANUS is augmented by a deep neural network that approximates molecular properties via active learning for enhanced sampling of the chemical space. Our method uses the SELFIES molecular representation and the STONED algorithm for the efficient generation of structures, and outperforms other generative models in common inverse molecular design tasks achieving state-of-the-art performance.

Abductive reasoning starts from some observations and aims at finding the most plausible explanation for these observations. To perform abduction, humans often make use of temporal and causal inferences, and knowledge about how some hypothetical situation can result in different outcomes. This work offers the first study of how such knowledge impacts the Abductive NLI task -- which consists in choosing the more likely explanation for given observations. We train a specialized language model LMI that is tasked to generate what could happen next from a hypothetical scenario that evolves from a given event. We then propose a multi-task model MTL to solve the Abductive NLI task, which predicts a plausible explanation by a) considering different possible events emerging from candidate hypotheses -- events generated by LMI -- and b) selecting the one that is most similar to the observed outcome. We show that our MTL model improves over prior vanilla pre-trained LMs fine-tuned on Abductive NLI. Our manual evaluation and analysis suggest that learning about possible next events from different hypothetical scenarios supports abductive inference.

Traditional automated theorem provers have relied on manually tuned heuristics to guide how they perform proof search. Recently, however, there has been a surge of interest in the design of learning mechanisms that can be integrated into theorem provers to improve their performance automatically. In this work, we introduce TRAIL, a deep learning-based approach to theorem proving that characterizes core elements of saturation-based theorem proving within a neural framework. TRAIL leverages (a) an effective graph neural network for representing logical formulas, (b) a novel neural representation of the state of a saturation-based theorem prover in terms of processed clauses and available actions, and (c) a novel representation of the inference selection process as an attention-based action policy. We show through a systematic analysis that these components allow TRAIL to significantly outperform previous reinforcement learning-based theorem provers on two standard benchmark datasets (up to 36% more theorems proved). In addition, to the best of our knowledge, TRAIL is the first reinforcement learning-based approach to exceed the performance of a state-of-the-art traditional theorem prover on a standard theorem proving benchmark (solving up to 17% more problems).

Given a stream of entries over time in a multi-aspect data setting where concept drift is present, how can we detect anomalous activities? Most of the existing unsupervised anomaly detection approaches seek to detect anomalous events in an offline fashion and require a large amount of data for training. This is not practical in real-life scenarios where we receive the data in a streaming manner and do not know the size of the stream beforehand. Thus, we need a data-efficient method that can detect and adapt to changing data trends, or concept drift, in an online manner. In this work, we propose MemStream, a streaming multi-aspect anomaly detection framework, allowing us to detect unusual events as they occur while being resilient to concept drift. We leverage the power of a denoising autoencoder to learn representations and a memory module to learn the dynamically changing trend in data without the need for labels. We prove the optimum memory size required for effective drift handling. Furthermore, MemStream makes use of two architecture design choices to be robust to memory poisoning. Experimental results show the effectiveness of our approach compared to state-of-the-art streaming baselines using 2 synthetic datasets and 11 real-world datasets.

In the new era of very large telescopes, where data is crucial to expand scientific knowledge, we have witnessed many deep learning applications for the automatic classification of lightcurves. Recurrent neural networks (RNNs) are one of the models used for these applications, and the LSTM unit stands out for being an excellent choice for the representation of long time series. In general, RNNs assume observations at discrete times, which may not suit the irregular sampling of lightcurves. A traditional technique to address irregular sequences consists of adding the sampling time to the network's input, but this is not guaranteed to capture sampling irregularities during training. Alternatively, the Phased LSTM unit has been created to address this problem by updating its state using the sampling times explicitly. In this work, we study the effectiveness of the LSTM and Phased LSTM based architectures for the classification of astronomical lightcurves. We use seven catalogs containing periodic and nonperiodic astronomical objects. Our findings show that LSTM outperformed PLSTM on 6/7 datasets. However, the combination of both units enhances the results in all datasets.

Automatic evaluation metrics are a crucial component of dialog systems research. Standard language evaluation metrics are known to be ineffective for evaluating dialog. As such, recent research has proposed a number of novel, dialog-specific metrics that correlate better with human judgements. Due to the fast pace of research, many of these metrics have been assessed on different datasets and there has as yet been no time for a systematic comparison between them. To this end, this paper provides a comprehensive assessment of recently proposed dialog evaluation metrics on a number of datasets. In this paper, 17 different automatic evaluation metrics are evaluated on 10 different datasets. Furthermore, the metrics are assessed in different settings, to better qualify their respective strengths and weaknesses. Metrics are assessed (1) on both the turn level and the dialog level, (2) for different dialog lengths, (3) for different dialog qualities (e.g., coherence, engaging), (4) for different types of response generation models (i.e., generative, retrieval, simple models and state-of-the-art models), (5) taking into account the similarity of different metrics and (6) exploring combinations of different metrics. This comprehensive assessment offers several takeaways pertaining to dialog evaluation metrics in general. It also suggests how to best assess evaluation metrics and indicates promising directions for future work.

In e-commerce advertising, it is crucial to jointly consider various performance metrics, e.g., user experience, advertiser utility, and platform revenue. Traditional auction mechanisms, such as GSP and VCG auctions, can be suboptimal due to their fixed allocation rules to optimize a single performance metric (e.g., revenue or social welfare). Recently, data-driven auctions, learned directly from auction outcomes to optimize multiple performance metrics, have attracted increasing research interests. However, the procedure of auction mechanisms involves various discrete calculation operations, making it challenging to be compatible with continuous optimization pipelines in machine learning. In this paper, we design \underline{D}eep \underline{N}eural \underline{A}uctions (DNAs) to enable end-to-end auction learning by proposing a differentiable model to relax the discrete sorting operation, a key component in auctions. We optimize the performance metrics by developing deep models to efficiently extract contexts from auctions, providing rich features for auction design. We further integrate the game theoretical conditions within the model design, to guarantee the stability of the auctions. DNAs have been successfully deployed in the e-commerce advertising system at Taobao. Experimental evaluation results on both large-scale data set as well as online A/B test demonstrated that DNAs significantly outperformed other mechanisms widely adopted in industry.