Many well-established recommender systems are based on representation learning in Euclidean space. In these models, matching functions such as the Euclidean distance or inner product are typically used for computing similarity scores between user and item embeddings. This paper investigates the notion of learning user and item representations in Hyperbolic space. In this paper, we argue that Hyperbolic space is more suitable for learning user-item embeddings in the recommendation domain. Unlike Euclidean spaces, Hyperbolic spaces are intrinsically equipped to handle hierarchical structure, encouraged by its property of exponentially increasing distances away from origin. We propose HyperBPR (Hyperbolic Bayesian Personalized Ranking), a conceptually simple but highly effective model for the task at hand. Our proposed HyperBPR not only outperforms their Euclidean counterparts, but also achieves state-of-the-art performance on multiple benchmark datasets, demonstrating the effectiveness of personalized recommendation in Hyperbolic space.

Quantized deep neural networks (QDNNs) are attractive due to their much lower memory storage and faster inference speed than their regular full precision counterparts. To maintain the same performance level especially at low bit-widths, QDNNs must be retrained. Their training involves piecewise constant activation functions and discrete weights, hence mathematical challenges arise. We introduce the notion of coarse derivative and propose the blended coarse gradient descent (BCGD) algorithm, for training fully quantized neural networks. Coarse gradient is generally not a gradient of any function but an artificial ascent direction. The weight update of BCGD goes by coarse gradient correction of a weighted average of the full precision weights and their quantization (the so-called blending), which yields sufficient descent in the objective value and thus accelerates the training. Our experiments demonstrate that this simple blending technique is very effective for quantization at extremely low bit-width such as binarization. In full quantization of ResNet-18 for ImageNet classification task, BCGD gives 64.36% top-1 accuracy with binary weights across all layers and 4-bit adaptive activation. If the weights in the first and last layers are kept in full precision, this number increases to 65.46%. As theoretical justification, we provide the convergence analysis of coarse gradient descent for a two-layer neural network model with Gaussian input data, and prove that the expected coarse gradient correlates positively with the underlying true gradient.

Random forest and deep neural network are two schools of effective classification methods in machine learning. While the random forest is robust irrespective of the data domain, the deep neural network has advantages in handling high dimensional data. In view that a differentiable neural decision forest can be added to the neural network to fully exploit the benefits of both models, in our work, we further combine convolutional autoencoder with neural decision forest, where autoencoder has its advantages in finding the hidden representations of the input data. We develop a gradient boost module and embed it into the proposed convolutional autoencoder with neural decision forest to improve the performance. The idea of gradient boost is to learn and use the residual in the prediction. In addition, we design a structure to learn the parameters of the neural decision forest and gradient boost module at contiguous steps. The extensive experiments on several public datasets demonstrate that our proposed model achieves good efficiency and prediction performance compared with a series of baseline methods.

The dominant, state-of-the-art collaborative filtering (CF) methods today mainly comprises neural models. In these models, deep neural networks, e.g.., multi-layered perceptrons (MLP), are often used to model nonlinear relationships between user and item representations. As opposed to shallow models (e.g., factorization-based models), deep models generally provide a greater extent of expressiveness, albeit at the expense of impaired/restricted information flow. Consequently, the performance of most neural CF models plateaus at 3-4 layers, with performance stagnating or even degrading when increasing the model depth. As such, the question of how to train really deep networks in the context of CF remains unclear. To this end, this paper proposes a new technique that enables training neural CF models all the way up to 20 layers and beyond. Our proposed approach utilizes a new hierarchical self-attention mechanism that learns introspective intra-feature similarity across all the hidden layers of a standard MLP model. All in all, our proposed architecture, SA-NCF (Self-Attentive Neural Collaborative Filtering) is a densely connected self-matching model that can be trained up to 24 layers without plateau-ing, achieving wide performance margins against its competitors. On several popular benchmark datasets, our proposed architecture achieves up to an absolute improvement of 23%-58% and 1.3x to 2.8x fold improvement in terms of nDCG@10 and Hit Ratio (HR@10) scores over several strong neural CF baselines.

Kernel methods are powerful tools to capture nonlinear patterns behind data. They implicitly learn high (even infinite) dimensional nonlinear features in the Reproducing Kernel Hilbert Space (RKHS) while making the computation tractable by leveraging the kernel trick. Classic kernel methods learn a single layer of nonlinear features, whose representational power may be limited. Motivated by recent success of deep neural networks (DNNs) that learn multi-layer hierarchical representations, we propose a Stacked Kernel Network (SKN) that learns a hierarchy of RKHS-based nonlinear features. SKN interleaves several layers of nonlinear transformations (from a linear space to a RKHS) and linear transformations (from a RKHS to a linear space). Similar to DNNs, a SKN is composed of multiple layers of hidden units, but each parameterized by a RKHS function rather than a finite-dimensional vector. We propose three ways to represent the RKHS functions in SKN: (1)nonparametric representation, (2)parametric representation and (3)random Fourier feature representation. Furthermore, we expand SKN into CNN architecture called Stacked Kernel Convolutional Network (SKCN). SKCN learning a hierarchy of RKHS-based nonlinear features by convolutional operation with each filter also parameterized by a RKHS function rather than a finite-dimensional matrix in CNN, which is suitable for image inputs. Experiments on various datasets demonstrate the effectiveness of SKN and SKCN, which outperform the competitive methods.