Group equivariant neural networks are used as building blocks of group invariant neural networks, which have been shown to improve generalisation performance and data efficiency through principled parameter sharing. Such works have mostly focused on group equivariant convolutions, building on the result that group equivariant linear maps are necessarily convolutions. In this work, we extend the scope of the literature to non-linear neural network modules, namely self-attention, that is emerging as a prominent building block of deep learning models. We propose the LieTransformer, an architecture composed of LieSelfAttention layers that are equivariant to arbitrary Lie groups and their discrete subgroups. We demonstrate the generality of our approach by showing experimental results that are competitive to baseline methods on a wide range of tasks: shape counting on point clouds, molecular property regression and modelling particle trajectories under Hamiltonian dynamics.

This research summary is just one of many that are distributed weekly on the AI scholar newsletter. To start receiving the weekly newsletter, sign up here. Artificial intelligence (AI) has grown tremendously in just a few years ushering us into the AI era. We now have self-driving cars, contemporary chatbots, high-end robots, recommender systems, advanced diagnostics systems, and more. Almost every research field is now using AI.

Neural networks(short for Artificial neural networks) model were inspired by the structure of neurons in our brain(biological neural networks). Each cell in a neural network is called a neuron and is connected to multiple neurons. Neurons in human (and mammalian) brains communicate by sending electrical signals between each other. But these are the only similarities between biological neural networks and artificial neural networks. A deep neural network is a specific type of neural network that excels at capturing nonlinear relationships in data.

Many interesting datasets ubiquitous in machine learning and deep learning can be described via graphs. As the scale and complexity of graph-structured datasets increase, such as in expansive social networks, protein folding, chemical interaction networks, and material phase transitions, improving the efficiency of the machine learning techniques applied to these is crucial. In this study, we focus on Graph Neural Networks (GNN), which have found great success in tasks such as node or edge classification and link prediction. However, standard GNN models have scaling limits due to necessary recursive calculations performed through dense graph relationships that lead to memory and runtime bottlenecks. While new approaches for processing larger networks are needed to advance graph techniques, and several have been proposed, we study how GNNs could be parallelized using existing tools and frameworks that are already known to be successful in the deep learning community. In particular, we investigate applying pipeline parallelism to GNN models with GPipe, introduced by Google in 2018.

In this paper, we consider the Click-Through-Rate (CTR) prediction problem. Factorization Machines and their variants consider pair-wise feature interactions, but normally we won't do high-order feature interactions using FM due to high time complexity. Given the success of deep neural networks (DNNs) in many fields, researchers have proposed several DNN-based models to learn high-order feature interactions. Multi-layer perceptrons (MLP) have been widely employed to learn reliable mappings from feature embeddings to final logits. In this paper, we aim to explore more about these high-order features interactions. However, high-order feature interaction deserves more attention and further development. Inspired by the great achievements of Densely Connected Convolutional Networks (DenseNet) in computer vision, we propose a novel model called Attentive DenseNet based Factorization Machines (AdnFM). AdnFM can extract more comprehensive deep features by using all the hidden layers from a feed-forward neural network as implicit high-order features, then selects dominant features via an attention mechanism. Also, high-order interactions in the implicit way using DNNs are more cost-efficient than in the explicit way, for example in FM. Extensive experiments on two real-world datasets show that the proposed model can effectively improve the performance of CTR prediction.

The drivers of compositionality in artificial languages that emerge when two (or more) agents play a non-visual referential game has been previously investigated using approaches based on the REINFORCE algorithm and the (Neural) Iterated Learning Model. Following the more recent introduction of the \textit{Straight-Through Gumbel-Softmax} (ST-GS) approach, this paper investigates to what extent the drivers of compositionality identified so far in the field apply in the ST-GS context and to what extent do they translate into (emergent) systematic generalisation abilities, when playing a visual referential game. Compositionality and the generalisation abilities of the emergent languages are assessed using topographic similarity and zero-shot compositional tests. Firstly, we provide evidence that the test-train split strategy significantly impacts the zero-shot compositional tests when dealing with visual stimuli, whilst it does not when dealing with symbolic ones. Secondly, empirical evidence shows that using the ST-GS approach with small batch sizes and an overcomplete communication channel improves compositionality in the emerging languages. Nevertheless, while shown robust with symbolic stimuli, the effect of the batch size is not so clear-cut when dealing with visual stimuli. Our results also show that not all overcomplete communication channels are created equal. Indeed, while increasing the maximum sentence length is found to be beneficial to further both compositionality and generalisation abilities, increasing the vocabulary size is found detrimental. Finally, a lack of correlation between the language compositionality at training-time and the agents' generalisation abilities is observed in the context of discriminative referential games with visual stimuli. This is similar to previous observations in the field using the generative variant with symbolic stimuli.

We present a novel deep reinforcement learning method to learn construction heuristics for vehicle routing problems. In specific, we propose a Multi-Decoder Attention Model (MDAM) to train multiple diverse policies, which effectively increases the chance of finding good solutions compared with existing methods that train only one policy. A customized beam search strategy is designed to fully exploit the diversity of MDAM. In addition, we propose an Embedding Glimpse layer in MDAM based on the recursive nature of construction, which can improve the quality of each policy by providing more informative embeddings. Extensive experiments on six different routing problems show that our method significantly outperforms the state-of-the-art deep learning based models.

Previous neural solvers of math word problems (MWPs) are learned with full supervision and fail to generate diverse solutions. In this paper, we address this issue by introducing a \textit{weakly-supervised} paradigm for learning MWPs. Our method only requires the annotations of the final answers and can generate various solutions for a single problem. To boost weakly-supervised learning, we propose a novel \textit{learning-by-fixing} (LBF) framework, which corrects the misperceptions of the neural network via symbolic reasoning. Specifically, for an incorrect solution tree generated by the neural network, the \textit{fixing} mechanism propagates the error from the root node to the leaf nodes and infers the most probable fix that can be executed to get the desired answer. To generate more diverse solutions, \textit{tree regularization} is applied to guide the efficient shrinkage and exploration of the solution space, and a \textit{memory buffer} is designed to track and save the discovered various fixes for each problem. Experimental results on the Math23K dataset show the proposed LBF framework significantly outperforms reinforcement learning baselines in weakly-supervised learning. Furthermore, it achieves comparable top-1 and much better top-3/5 answer accuracies than fully-supervised methods, demonstrating its strength in producing diverse solutions.

Deep learning (DL) models are being deployed at medical centers to aid radiologists for diagnosis of lung conditions from chest radiographs. Such models are often trained on a large volume of publicly available labeled radiographs. These pre-trained DL models' ability to generalize in clinical settings is poor because of the changes in data distributions between publicly available and privately held radiographs. In chest radiographs, the heterogeneity in distributions arises from the diverse conditions in X-ray equipment and their configurations used for generating the images. In the machine learning community, the challenges posed by the heterogeneity in the data generation source is known as domain shift, which is a mode shift in the generative model. In this work, we introduce a domain-shift detection and removal method to overcome this problem. Our experimental results show the proposed method's effectiveness in deploying a pre-trained DL model for abnormality detection in chest radiographs in a clinical setting.

Video analytics applications use edge compute servers for the analytics of the videos (for bandwidth and privacy). Compressed models that are deployed on the edge servers for inference suffer from data drift, where the live video data diverges from the training data. Continuous learning handles data drift by periodically retraining the models on new data. Our work addresses the challenge of jointly supporting inference and retraining tasks on edge servers, which requires navigating the fundamental tradeoff between the retrained model's accuracy and the inference accuracy. Our solution Ekya balances this tradeoff across multiple models and uses a micro-profiler to identify the models that will benefit the most by retraining. Ekya's accuracy gain compared to a baseline scheduler is 29% higher, and the baseline requires 4x more GPU resources to achieve the same accuracy as Ekya.