Natural language inference (NLI) requires models to learn and apply commonsense knowledge. These reasoning abilities are particularly important for explainable NLI systems that generate a natural language explanation in addition to their label predic
tion. The integration of external knowledge has been shown to improve NLI systems, here we investigate whether it can also improve their explanation capabilities. For this, we investigate different sources of external knowledge and evaluate the performance of our models on in-domain data as well as on special transfer datasets that are designed to assess fine-grained reasoning capabilities. We find that different sources of knowledge have a different effect on reasoning abilities, for example, implicit knowledge stored in language models can hinder reasoning on numbers and negations. Finally, we conduct the largest and most fine-grained explainable NLI crowdsourcing study to date. It reveals that even large differences in automatic performance scores do neither reflect in human ratings of label, explanation, commonsense nor grammar correctness.
We introduce ROS-X-Habitat, a software interface that bridges the AI Habitat platform for embodied reinforcement learning agents with other robotics resources via ROS. This interface not only offers standardized communication protocols between embodi
ed agents and simulators, but also enables physics-based simulation. With this interface, roboticists are able to train their own Habitat RL agents in another simulation environment or to develop their own robotic algorithms inside Habitat Sim. Through in silico experiments, we demonstrate that ROS-X-Habitat has minimal impact on the navigation performance and simulation speed of Habitat agents; that a standard set of ROS mapping, planning and navigation tools can run in the Habitat simulator, and that a Habitat agent can run in the standard ROS simulator Gazebo.
Differentiable neural architecture search (DNAS) is known for its capacity in the automatic generation of superior neural networks. However, DNAS based methods suffer from memory usage explosion when the search space expands, which may prevent them f
rom running successfully on even advanced GPU platforms. On the other hand, reinforcement learning (RL) based methods, while being memory efficient, are extremely time-consuming. Combining the advantages of both types of methods, this paper presents RADARS, a scalable RL-aided DNAS framework that can explore large search spaces in a fast and memory-efficient manner. RADARS iteratively applies RL to prune undesired architecture candidates and identifies a promising subspace to carry out DNAS. Experiments using a workstation with 12 GB GPU memory show that on CIFAR-10 and ImageNet datasets, RADARS can achieve up to 3.41% higher accuracy with 2.5X search time reduction compared with a state-of-the-art RL-based method, while the two DNAS baselines cannot complete due to excessive memory usage or search time. To the best of the authors knowledge, this is the first DNAS framework that can handle large search spaces with bounded memory usage.
We investigate joint source channel coding (JSCC) for wireless image transmission over multipath fading channels. Inspired by recent works on deep learning based JSCC and model-based learning methods, we combine an autoencoder with orthogonal frequen
cy division multiplexing (OFDM) to cope with multipath fading. The proposed encoder and decoder use convolutional neural networks (CNNs) and directly map the source images to complex-valued baseband samples for OFDM transmission. The multipath channel and OFDM are represented by non-trainable (deterministic) but differentiable layers so that the system can be trained end-to-end. Furthermore, our JSCC decoder further incorporates explicit channel estimation, equalization, and additional subnets to enhance the performance. The proposed method exhibits 2.5 -- 4 dB SNR gain for the equivalent image quality compared to conventional schemes that employ state-of-the-art but separate source and channel coding such as BPG and LDPC. The performance further improves when the system incorporates the channel state information (CSI) feedback. The proposed scheme is robust against OFDM signal clipping and parameter mismatch for the channel model used in training and evaluation.
While recent research on natural language inference has considerably benefited from large annotated datasets, the amount of inference-related knowledge (including commonsense) provided in the annotated data is still rather limited. There have been tw
o lines of approaches that can be used to further address the limitation: (1) unsupervised pretraining can leverage knowledge in much larger unstructured text data; (2) structured (often human-curated) knowledge has started to be considered in neural-network-based models for NLI. An immediate question is whether these two approaches complement each other, or how to develop models that can bring together their advantages. In this paper, we propose models that leverage structured knowledge in different components of pre-trained models. Our results show that the proposed models perform better than previous BERT-based state-of-the-art models. Although our models are proposed for NLI, they can be easily extended to other sentence or sentence-pair classification problems.
A new strategy, namely the clean numerical simulation (CNS), was proposed (J. Computational Physics, 418:109629, 2020) to gain reliable/convergent simulations (with negligible numerical noises) of spatiotemporal chaotic systems in a long enough inter
val of time, which provide us benchmark solution for comparison. Here we illustrate that machine learning (ML) can always give good enough fitting predictions of a spatiotemporal chaos by using, separately, two quite different training sets: one is the clean database given by the CNS with negligible numerical noises, the other is the polluted database given by the traditional algorithms in single/double precision with considerably large numerical noises. However, even in statistics, the ML predictions based on the polluted database are quite different from those based on the clean database. It illustrates that the database noises have huge influences on ML predictions of some spatiotemporal chaos, even in statistics. Thus, we must use a clean database for machine learning of some spatiotemporal chaos. This surprising result might open a new door and possibility to study machine learning.
RGBD (RGB plus depth) object tracking is gaining momentum as RGBD sensors have become popular in many application fields such as robotics.However, the best RGBD trackers are extensions of the state-of-the-art deep RGB trackers. They are trained with
RGB data and the depth channel is used as a sidekick for subtleties such as occlusion detection. This can be explained by the fact that there are no sufficiently large RGBD datasets to 1) train deep depth trackers and to 2) challenge RGB trackers with sequences for which the depth cue is essential. This work introduces a new RGBD tracking dataset - Depth-Track - that has twice as many sequences (200) and scene types (40) than in the largest existing dataset, and three times more objects (90). In addition, the average length of the sequences (1473), the number of deformable objects (16) and the number of annotated tracking attributes (15) have been increased. Furthermore, by running the SotA RGB and RGBD trackers on DepthTrack, we propose a new RGBD tracking baseline, namely DeT, which reveals that deep RGBD tracking indeed benefits from genuine training data. The code and dataset is available at https://github.com/xiaozai/DeT
Sampling is a critical operation in the training of Graph Neural Network (GNN) that helps reduce the cost. Previous works have explored improving sampling algorithms through mathematical and statistical methods. However, there is a gap between sampli
ng algorithms and hardware. Without consideration of hardware, algorithm designers merely optimize sampling at the algorithm level, missing the great potential of promoting the efficiency of existing sampling algorithms by leveraging hardware features. In this paper, we first propose a unified programming model for mainstream sampling algorithms, termed GNNSampler, covering the key processes for sampling algorithms in various categories. Second, we explore the data locality among nodes and their neighbors (i.e., the hardware feature) in real-world datasets for alleviating the irregular memory access in sampling. Third, we implement locality-aware optimizations in GNNSampler for diverse sampling algorithms to optimize the general sampling process in the training of GNN. Finally, we emphatically conduct experiments on large graph datasets to analyze the relevance between the training time, model accuracy, and hardware-level metrics, which helps achieve a good trade-off between time and accuracy in GNN training. Extensive experimental results show that our method is universal to mainstream sampling algorithms and reduces the training time of GNN (range from 4.83% with layer-wise sampling to 44.92% with subgraph-based sampling) with comparable accuracy.
Access to labeled time series data is often limited in the real world, which constrains the performance of deep learning models in the field of time series analysis. Data augmentation is an effective way to solve the problem of small sample size and
imbalance in time series datasets. The two key factors of data augmentation are the distance metric and the choice of interpolation method. SMOTE does not perform well on time series data because it uses a Euclidean distance metric and interpolates directly on the object. Therefore, we propose a DTW-based synthetic minority oversampling technique using siamese encoder for interpolation named DTWSSE. In order to reasonably measure the distance of the time series, DTW, which has been verified to be an effective method forts, is employed as the distance metric. To adapt the DTW metric, we use an autoencoder trained in an unsupervised self-training manner for interpolation. The encoder is a Siamese Neural Network for mapping the time series data from the DTW hidden space to the Euclidean deep feature space, and the decoder is used to map the deep feature space back to the DTW hidden space. We validate the proposed methods on a number of different balanced or unbalanced time series datasets. Experimental results show that the proposed method can lead to better performance of the downstream deep learning model.
In this paper, we performed thermodynamic and electron spin resonance (ESR) measurements to study low-energy magnetic excitations, which were significantly affected by crystalline electric field (CEF) excitations due to relatively small gaps between
the CEF ground state and the excited states. Based on the CEF and mean-field (MF) theories, we analyzed systematically and consistently the ESR experiments and thermodynamic measurements including susceptibility, magnetization, and heat capacity. The CEF parameters were successfully extracted by fitting high-temperature (> 20 K) susceptibilities in the ab-plane and along the c-axis, allowing to determine the Lande factors ($g_{ab,calc}$ = 5.98(7) and $g_{c,calc}$ = 2.73(3)). These values were consistent with the values of Lande factors determined by ESR experiments ($g_{ab,exp}$ = 5.69 and $g_{c,exp}$ = 2.75). By applying the CEF and MF theories to the susceptibility and magnetization results, we estimated the anisotropic spin-exchange energies and found that the CEF excitations in ce{KErTe2} played a decisive role in the magnetism above 3 K, while the low-temperature magnetism below 10 K was gradually correlated with the anisotropic spin-exchange interactions. The CEF excitations were demonstrated in the low-temperature heat capacity, where both the positions of two broad peaks and their magnetic field dependence well corroborated our calculations. The present study provides a basis to explore the enriched magnetic and electronic properties of the QSL family.
