This survey investigates current techniques for representing qualitative data for use as input to neural networks. Techniques for using qualitative data in neural networks are well known. However, researchers continue to discover new variations or entirely new methods for working with categorical data in neural networks. Our primary contribution is to cover these representation techniques in a single work. Practitioners working with big data often have a need to encode categorical values in their datasets in order to leverage machine learning algorithms. Moreover, the size of data sets we consider as big data may cause one to reject some encoding techniques as impractical, due to their running time complexity. Neural networks take vectors of real numbers as inputs. One must use a technique to map qualitative values to numerical values before using them as input to a neural network. These techniques are known as embeddings, encodings, representations, or distributed representations. Another contribution this work makes is to provide references for the source code of various techniques, where we are able to verify the authenticity of the source code. We cover recent research in several domains where researchers use categorical data in neural networks. Some of these domains are natural language processing, fraud detection, and clinical document automation. This study provides a starting point for research in determining which techniques for preparing qualitative data for use with neural networks are best. It is our intention that the reader should use these implementations as a starting point to design experiments to evaluate various techniques for working with qualitative data in neural networks. The third contribution we make in this work is a new perspective on techniques for using categorical data in neural networks. We organize techniques for using categorical data in neural networks into three categories. We find three distinct patterns in techniques that identify a technique as determined, algorithmic, or automated. The fourth contribution we make is to identify several opportunities for future research. The form of the data that one uses as an input to a neural network is crucial for using neural networks effectively. This work is a tool for researchers to find the most effective technique for working with categorical data in neural networks, in big data settings. To the best of our knowledge this is the first in-depth look at techniques for working with categorical data in neural networks.

/pdf/survey-on-categorical-data-for-neural-networks-28h02g6ago.pdf

Survey on categorical data for neural networks

Machine learning techniques have recently been adopted in various applications in medicine, biology, chemistry, and material engineering. An important task is to predict the properties of molecules, which serves as the main subroutine in many downstream applications such as virtual screening and drug design. Despite the increasing interest, the key challenge is to construct proper representations of molecules for learning algorithms. This paper introduces the N-gram graph, a simple unsupervised representation for molecules. The method first embeds the vertices in the molecule graph. It then constructs a compact representation for the graph by assembling the vertex embeddings in short walks in the graph, which we show is equivalent to a simple graph neural network that needs no training. The representations can thus be efficiently computed and then used with supervised learning methods for prediction. Experiments on 60 tasks from 10 benchmark datasets demonstrate its advantages over both popular graph neural networks and traditional representation methods. This is complemented by theoretical analysis showing its strong representation and prediction power.

/pdf/n-gram-graph-simple-unsupervised-representation-for-graphs-4y363y5krn.pdf

N-Gram Graph: Simple Unsupervised Representation for Graphs, with Applications to Molecules

With the increasing computational demands of the neural networks, many hardware accelerators for the neural networks have been proposed. Such existing neural network accelerators often focus on popular neural network types such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs); however, not much attention has been paid to attention mechanisms, an emerging neural network primitive that enables neural networks to retrieve most relevant information from a knowledge-base, external memory, or past states. The attention mechanism is widely adopted by many state-of-the-art neural networks for computer vision, natural language processing, and machine translation, and accounts for a large portion of total execution time. We observe today's practice of implementing this mechanism using matrix-vector multiplication is suboptimal as the attention mechanism is semantically a content-based search where a large portion of computations ends up not being used. Based on this observation, we design and architect A3, which accelerates attention mechanisms in neural networks with algorithmic approximation and hardware specialization. Our proposed accelerator achieves multiple orders of magnitude improvement in energy efficiency (performance/watt) as well as substantial speedup over the state-of-the-art conventional hardware.

A^3: Accelerating Attention Mechanisms in Neural Networks with Approximation

Modern deep learning-based recommendation systems exploit hundreds to thousands of different categorical features, each with millions of different categories ranging from clicks to posts. To respect the natural diversity within the categorical data, embeddings map each category to a unique dense representation within an embedded space. Since each categorical feature could take on as many as tens of millions of different possible categories, the embedding tables form the primary memory bottleneck during both training and inference. We propose a novel approach for reducing the embedding size in an end-to-end fashion by exploiting complementary partitions of the category set to produce a unique embedding vector for each category without explicit definition. By storing multiple smaller embedding tables based on each complementary partition and combining embeddings from each table, we define a unique embedding for each category at smaller cost. This approach may be interpreted as using a specific fixed codebook to ensure uniqueness of each category's representation. Our experimental results demonstrate the effectiveness of our approach over the hashing trick for reducing the size of the embedding tables in terms of model loss and accuracy, while retaining a similar reduction in the number of parameters.

Compositional Embeddings Using Complementary Partitions for Memory-Efficient Recommendation Systems

Pre-trained word embeddings are used in several downstream applications as well as for constructing representations for sentences, paragraphs and documents. Recently, there has been an emphasis on improving the pretrained word vectors through post-processing algorithms. One improvement area is reducing the dimensionality of word embeddings. Reducing the size of word embeddings can improve their utility in memory constrained devices, benefiting several real world applications. In this work, we present a novel technique that efficiently combines PCA based dimensionality reduction with a recently proposed post-processing algorithm (Mu and Viswanath, 2018), to construct effective word embeddings of lower dimensions. Empirical evaluations on several benchmarks show that our algorithm efficiently reduces the embedding size while achieving similar or (more often) better performance than original embeddings. We have released the source code along with this paper.

/pdf/effective-dimensionality-reduction-for-word-embeddings-1ht9cklr8o.pdf

Effective Dimensionality Reduction for Word Embeddings.

In this paper, we provide a theoretical understanding of word embedding and its dimensionality. Motivated by the unitary-invariance of word embedding, we propose the Pairwise Inner Product (PIP) loss, a novel metric on the dissimilarity between word embeddings. Using techniques from matrix perturbation theory, we reveal a fundamental bias-variance trade-off in dimensionality selection for word embeddings. This bias-variance trade-off sheds light on many empirical observations which were previously unexplained, for example the existence of an optimal dimensionality. Moreover, new insights and discoveries, like when and how word embeddings are robust to over-fitting, are revealed. By optimizing over the bias-variance trade-off of the PIP loss, we can explicitly answer the open question of dimensionality selection for word embedding.

/pdf/on-the-dimensionality-of-word-embedding-28dq22okqm.pdf

On the dimensionality of word embedding

On the Dimensionality of Word Embedding

In this paper, we present a new data mining framework for discovering sequence effects. In particular, we focus on the sequences consisting of actions that are taken in chronological order, like sequences of clinical procedures or marketing actions. Each sequence is associated with a binary outcome, a success or a failure. We investigate the hypothesis that certain subsequences of actions contribute to successes, which we call effective subsequences. A generic data mining algorithm for extracting effective subsequences is proposed, which is verified both quantitatively and qualitatively. We experimented our effective subsequence mining algorithm on a real sales opportunity dataset. Based on the subsequence model, a market campaign attribution model is proposed, with application to the same sales dataset.

Yuanyuan Shen

Papers

On the dimensionality of word embedding

On the Dimensionality of Word Embedding

Mining Effective Subsequences with Application in Marketing Attribution