This paper addresses deep face recognition (FR) problem under open-set protocol, where ideal features are expected to have smaller maximal intra-class distance than minimal inter-class a suitably chosen metric space. However, few existing algorithms can effectively achieve this criterion. To end, we propose the angular softmax (A-Softmax) loss that enables convolutional neural networks (CNNs) learn angularly discriminative features. Geometrically, A-Softmax be viewed as imposing constraints...

The design of good heuristics or approximation algorithms for NP-hard combinatorial optimization problems often requires significant specialized knowledge and trial-and-error. Can we automate this challenging, tedious process, learn the instead? In many real-world applications, it is typically case that same problem solved again on a regular basis, maintaining structure but differing in data. This provides an opportunity learning heuristic exploit such recurring problems. paper, propose...

Deep learning methods exhibit promising performance for predictive modeling in healthcare, but two important challenges remain: -

The problem of cross-platform binary code similarity detection aims at detecting whether two functions coming from different platforms are similar or not. It has many security applications, including plagiarism detection, malware vulnerability search, etc. Existing approaches rely on approximate graph matching algorithms, which inevitably slow and sometimes inaccurate, hard to adapt a new task. To address these issues, in this work, we propose novel neural network-based approach compute the...

Kernel classifiers and regressors designed for structured data, such as sequences, trees graphs, have significantly advanced a number of interdisciplinary areas computational biology drug design. Typically, kernels are beforehand data type which either exploit statistics the structures or make use probabilistic generative models, then discriminative classifier is learned based on via convex optimization. However, an elegant two-stage approach also limited kernel methods from scaling up to...

We introduce a framework for filtering features that employs the Hilbert-Schmidt Independence Criterion (HSIC) as measure of dependence between and labels.The key idea is good should maximise such dependence.Feature selection various supervised learning problems (including classification regression) unified under this framework, solutions can be approximated using backward-elimination algorithm.We demonstrate usefulness our method on both artificial real world datasets.

Knowledge graph (KG) is known to be helpful for the task of question answering (QA), since it provides well-structured relational information between entities, and allows one further infer indirect facts. However, challenging build QA systems which can learn reason over knowledge graphs based on question-answer pairs alone. First, when people ask questions, their expressions are noisy (for example, typos in texts, or variations pronunciations), non-trivial system match those mentioned...

Large-scale datasets possessing clean label annotations are crucial for training Convolutional Neural Networks (CNNs). However, labeling large-scale data can be very costly and error-prone, even high-quality likely to contain noisy (incorrect) labels. Existing works usually employ a closed-set assumption, whereby the samples associated with labels possess true class contained within set of known classes in data. such an assumption is too restrictive many applications, since might fact that...

We introduce instancewise feature selection as a methodology for model interpretation. Our method is based on learning function to extract subset of features that are most informative each given example. This selector trained maximize the mutual information between selected and response variable, where conditional distribution variable input be explained. develop an efficient variational approximation information, show effectiveness our variety synthetic real data sets using both...

Deep learning on graph structures has shown exciting results in various applications. However, few attentions have been paid to the robustness of such models, contrast numerous research work for image or text adversarial attack and defense. In this paper, we focus attacks that fool model by modifying combinatorial structure data. We first propose a reinforcement based method learns generalizable policy, while only requiring prediction labels from target classifier. Also, variants genetic...

We present, GeniePath, a scalable approach for learning adaptive receptive fields of neural networks defined on permutation invariant graph data. In we propose an path layer consists two complementary functions designed breadth and depth exploration respectively, where the former learns importance different sized neighborhoods, while latter extracts filters signals aggregated from neighbors hops away. Our method works in both transductive inductive settings, extensive experiments compared...

This paper addresses deep face recognition (FR) problem under open-set protocol, where ideal features are expected to have smaller maximal intra-class distance than minimal inter-class a suitably chosen metric space. However, few existing algorithms can effectively achieve this criterion. To end, we propose the angular softmax (A-Softmax) loss that enables convolutional neural networks (CNNs) learn angularly discriminative features. Geometrically, A-Softmax be viewed as imposing constraints...

The design of strategies for branching in Mixed Integer Programming (MIP) is guided by cycles parameter tuning and offline experimentation on an extremely heterogeneous testbed, using the average performance. Once devised, these (and their settings) are essentially input-agnostic. To address issues, we propose a machine learning (ML) framework variable MIP.Our method observes decisions made Strong Branching (SB), time-consuming strategy that produces small search trees, collecting features...

In this paper, we extend the Hilbert space embedding approach to handle conditional distributions. We derive a kernel estimate for embedding, and show its connection ordinary embeddings. Conditional embeddings largely our ability manipulate distributions in spaces, as an example, nonparametric method modeling dynamical systems where belief state of system is maintained embedding. Our very general terms both domains types that it can handle, demonstrate effectiveness various systems. expect...

The fully-connected layers of deep convolutional neural networks typically contain over 90% the network parameters. Reducing number parameters while preserving predictive performance is critically important for training big models in distributed systems and deployment embedded devices. In this paper, we introduce a novel Adaptive Fastfood transform to reparameterize matrix-vector multiplication fully connected layers. Reparameterizing layer with d inputs n outputs reduces storage...

The availability of large scale event data with time stamps has given rise to dynamically evolving knowledge graphs that contain temporal information for each edge. Reasoning over in such dynamic is not yet well understood. To this end, we present Know-Evolve, a novel deep evolutionary network learns non-linearly entity representations time. occurrence fact (edge) modeled as multivariate point process whose intensity function modulated by the score computed based on learned embeddings. We...

Deep generative models have been enjoying success in modeling continuous data. However it remains challenging to capture the representations for discrete structures with formal grammars and semantics, e.g., computer programs molecular structures. How generate both syntactically semantically correct data still largely an open problem. Inspired by theory of compiler where syntax semantics check is done via syntax-directed translation (SDT), we propose a novel variational autoencoder (SD-VAE)...

Knowledge graph (KG) is known to be helpful for the task of question answering (QA), since it provides well-structured relational information between entities, and allows one further infer indirect facts. However, challenging build QA systems which can learn reason over knowledge graphs based on question-answer pairs alone. First, when people ask questions, their expressions are noisy (for example, typos in texts, or variations pronunciations), non-trivial system match those mentioned...

Many modern applications of signal processing and machine learning, ranging from computer vision to computational biology, require the analysis large volumes high-dimensional continuous-valued measurements. Complex statistical features are commonplace, including multimodality, skewness, rich dependency structures. Such problems call for a flexible robust modeling framework that can take into account these diverse features. Most existing approaches, graphical models, rely heavily on...