Zhang, Jie

Spectral Network Embedding: A Fast and Scalable Method via Sparsity

arXiv.org Artificial Intelligence

Network embedding aims to learn low-dimensional representations of nodes in a network, while the network structure and inherent properties are preserved. It has attracted tremendous attention recently due to significant progress in downstream network learning tasks, such as node classification, link prediction, and visualization. However, most existing network embedding methods suffer from the expensive computations due to the large volume of networks. In this paper, we propose a $10\times \sim 100\times$ faster network embedding method, called Progle, by elegantly utilizing the sparsity property of online networks and spectral analysis. In Progle, we first construct a \textit{sparse} proximity matrix and train the network embedding efficiently via sparse matrix decomposition. Then we introduce a network propagation pattern via spectral analysis to incorporate local and global structure information into the embedding. Besides, this model can be generalized to integrate network information into other insufficiently trained embeddings at speed. Benefiting from sparse spectral network embedding, our experiment on four different datasets shows that Progle outperforms or is comparable to state-of-the-art unsupervised comparison approaches---DeepWalk, LINE, node2vec, GraRep, and HOPE, regarding accuracy, while is $10\times$ faster than the fastest word2vec-based method. Finally, we validate the scalability of Progle both in real large-scale networks and multiple scales of synthetic networks.

Dynamically Hierarchy Revolution: DirNet for Compressing Recurrent Neural Network on Mobile Devices

arXiv.org Machine Learning

Recurrent neural networks (RNNs) achieve cutting-edge performance on a variety of problems. However, due to their high computational and memory demands, deploying RNNs on resource constrained mobile devices is a challenging task. To guarantee minimum accuracy loss with higher compression rate and driven by the mobile resource requirement, we introduce a novel model compression approach DirNet based on an optimized fast dictionary learning algorithm, which 1) dynamically mines the dictionary atoms of the projection dictionary matrix within layer to adjust the compression rate 2) adaptively changes the sparsity of sparse codes cross the hierarchical layers. Experimental results on language model and an ASR model trained with a 1000h speech dataset demonstrate that our method significantly outperforms prior approaches. Evaluated on off-the-shelf mobile devices, we are able to reduce the size of original model by eight times with real-time model inference and negligible accuracy loss.

An Unsupervised Clustering-Based Short-Term Solar Forecasting Methodology Using Multi-Model Machine Learning Blending

arXiv.org Machine Learning

Solar forecasting accuracy is affected by weather conditions, and weather awareness forecasting models are expected to improve the performance. However, it may not be available and reliable to classify different forecasting tasks by using only meteorological weather categorization. In this paper, an unsupervised clustering-based (UC-based) solar forecasting methodology is developed for short-term (1-hour-ahead) global horizontal irradiance (GHI) forecasting. This methodology consists of three parts: GHI time series unsupervised clustering, pattern recognition, and UC-based forecasting. The daily GHI time series is first clustered by an Optimized Cross-validated ClUsteRing (OCCUR) method, which determines the optimal number of clusters and best clustering results. Then, support vector machine pattern recognition (SVM-PR) is adopted to recognize the category of a certain day using the first few hours' data in the forecasting stage. GHI forecasts are generated by the most suitable models in different clusters, which are built by a two-layer Machine learning based Multi-Model (M3) forecasting framework. The developed UC-based methodology is validated by using 1-year of data with six solar features. Numerical results show that (i) UC-based models outperform non-UC (all-in-one) models with the same M3 architecture by approximately 20%; (ii) M3-based models also outperform the single-algorithm machine learning (SAML) models by approximately 20%.

Hourly-Similarity Based Solar Forecasting Using Multi-Model Machine Learning Blending

arXiv.org Machine Learning

With the increasing penetration of solar power into power systems, forecasting becomes critical in power system operations. In this paper, an hourly-similarity (HS) based method is developed for 1-hour-ahead (1HA) global horizontal irradiance (GHI) forecasting. This developed method utilizes diurnal patterns, statistical distinctions between different hours, and hourly similarities in solar data to improve the forecasting accuracy. The HS-based method is built by training multiple two-layer multi-model forecasting framework (MMFF) models independently with the same-hour subsets. The final optimal model is a combination of MMFF models with the best-performed blending algorithm at every hour. At the forecasting stage, the most suitable model is selected to perform the forecasting subtask of a certain hour. The HS-based method is validated by 1-year data with six solar features collected by the National Renewable Energy Laboratory (NREL). Results show that the HS-based method outperforms the non-HS (all-in-one) method significantly with the same MMFF architecture, wherein the optimal HS- based method outperforms the best all-in-one method by 10.94% and 7.74% based on the normalized mean absolute error and normalized root mean square error, respectively.

Risk-Aware Proactive Scheduling via Conditional Value-at-Risk

AAAI Conferences

In this paper, we consider the challenging problem of riskaware proactive scheduling with the objective of minimizing robust makespan. State-of-the-art approaches based on probabilistic constrained optimization lead to Mixed Integer Linear Programs that must be heuristically approximated. We optimize the robust makespan via a coherent risk measure, Conditional Value-at-Risk (CVaR). Since traditional CVaR optimization approaches assuming linear spaces does not suit our problem, we propose a general branch-and-bound framework for combinatorial CVaR minimization. We then design an approximate complete algorithm, and employ resource reasoning to enable constraint propagation for multiple samples. Empirical results show that our algorithm outperforms state-of-the-art approaches with higher solution quality.

Average-Case Approximation Ratio of Scheduling Without Payments

AAAI Conferences

Apart from the principles and methodologies inherited from Economics and Game Theory, the studies in Algorithmic Mechanism Design typically employ the worst-case analysis and approximation schemes of Theoretical Computer Science. For instance, the approximation ratio, which is the canonical measure of evaluating how well an incentive-compatible mechanism approximately optimizes the objective, is defined in the worst-case sense. It compares the performance of the optimal mechanism against the performance of a truthful mechanism, for all possible inputs. In this paper, we take the average-case analysis approach, and tackle one of the primary motivating problems in Algorithmic Mechanism Design -- the scheduling problem [Nisan and Ronen 1999]. One version of this problem which includes a verification component is studied by [Koutsoupias 2014]. It was shown that the problem has a tight approximation ratio bound of (n+1)/2 for the single-task setting, where n is the number of machines. We show, however, when the costs of the machines to executing the task follow any independent and identical distribution, the average-case approximation ratio of the mechanism given in [Koutsoupias 2014] is upper bounded by a constant. This positive result asymptotically separates the average-case ratio from the worst-case ratio, and indicates that the optimal mechanism for the problem actually works well on average, although in the worst-case the expected cost of the mechanism is Theta(n) times that of the optimal cost.

FILE: A Novel Framework for Predicting Social Status in Signed Networks

AAAI Conferences

Link prediction in signed social networks is challenging because of the existence and imbalance of the three kinds of social status (positive, negative and no-relation). Furthermore, there are a variety types of no-relation status in reality, e.g., strangers and frenemies, which cannot be well distinguished from the other linked status by existing approaches. In this paper, we propose a novel Framework of Integrating both Latent and Explicit features (FILE), to better deal with the no-relation status and improve the overall link prediction performance in signed networks. In particular, we design two latent features from latent space and two explicit features by extending social theories, and learn these features for each user via matrix factorization with a specially designed ranking-oriented loss function. Experimental results demonstrate the superior of our approach over state-of-the-art methods.

POMDP-Based Decision Making for Fast Event Handling in VANETs

AAAI Conferences

Malicious vehicle agents broadcast fake information about traffic events and thereby undermine the benefits of vehicle-to-vehicle communication in vehicular ad-hoc networks (VANETs). Trust management schemes addressing this issue do not focus on effective/fast decision making in reacting to traffic events. We propose a Partially Observable Markov Decision Process (POMDP) based approach to balance the trade-off between information gathering and exploiting actions resulting in faster responses. Our model copes with malicious behavior by maintaining it as part of a small state space, thus is scalable for large VANETs. We also propose an algorithm to learn model parameters in a dynamic behavior setting. Experimental results demonstrate that our model can effectively balance the decision quality and response time while still being robust to sophisticated malicious attacks.

Rethinking the Link Prediction Problem in Signed Social Networks

AAAI Conferences

We rethink the link prediction problem in signed social networks by also considering "no-relation" as a future status of a node pair, rather than simply distinguishing positive and negative links proposed in the literature. To understand the underlying mechanism of link formation in signed networks, we propose a feature framework on the basis of a thorough exploration of potential features for the newly identified problem. Grounded on the framework, we also design a trinary classification model, and experimental results show that our method outperforms the state-of-the-art approaches.