Andrzej Szwabe

Semantically enhanced collaborative filtering based on RSVD

We investigate a hybrid recommendation method that is based on two-stage data processing - first dealing with content features describing items, and then handling user behavioral data. The evaluation of the proposed method is oriented on the so-called find-good-items task, rather than on the low-error-of-ratings prediction. We focus on a case of extreme collaborative data sparsity. Our method is a combination of content features preprocessing performed by means of Random Indexing (RI), a reflective retraining of preliminary reduced item vectors according to collaborative filtering data, and vector space optimization based on Singular Value Decomposition (SVD). We demonstrate that such an approach is appropriate in high data sparsity scenarios, which disqualify the use of widely-referenced collaborative filtering methods, and allows to generate more accurate recommendations than those obtained through a hybrid method based on weighted feature combination. Moreover, the proposed solution allows to improve the recommendation accuracy without increasing the computational complexity.

RSVD-based Dimensionality Reduction for Recommender Systems

We investigate dimensionality reduction methods from the perspective of their ability to produce a low-rank customer-product matrix representation. We analyze the results of using collaborative filtering based on SVD, RI, Reflective Random Indexing (RRI) and Randomized Singular Value Decomposition (RSVD) from the perspective of selected algebraic (i.e. application-independent) properties. We show that the Frobenius-norm optimality of SVD does not correspond to the optimal recommendation accuracy, when measured in terms of F1. On the other hand, a high collaborative filtering quality is achievable when a matrix decomposition - based on a combination of RRI and SVD referred to as RSVD-RRI - leads to increased diversity of low-dimensional eigenvectors. The diversity is observable from the perspective of cosine similarities analyzed in comparison to the analogical case of SVD. Such a feature is more desirable than the fidelity of the input matrix spectrum representation, despite the MSE-optimality of SVD.

Implementation of backpressure-based routing integrated with Max-Weight Scheduling in a wireless multi-hop network

Reliable transmission is one of the key objectives of research on wireless network optimization. The backpressure-based Max-Weight Scheduling (MWS) policy is theoretically proven to be the optimal solution for achieving the highest available throughput. On the other hand, proactive routing protocols, such as Optimized Link State Protocol (OLSR), are able to compute reliable paths, which, in turn, can be used as a basis for MWS-based resource allocation. However, in its standard implementation, OLSR is a single-path protocol, whereas it is known that at least in some scenarios MWS algorithms can provide better network performance when used together with multi-path packet forwarding. Following this motivation, we implemented a modification of the OLSR protocol aimed at realizing multi-path routing in cooperation with the backpressure policy. The scheduling component was implemented at the application layer, and was supported by a mechanism for an indirect estimation of the MAC-layer queue state. Thanks to this approach, the system may work with the unmodified existing 802.11 MAC, and can be deployed in existing wireless networks. We tested the proposed solutions in a realistic wireless network scenario.

Integration of Multi-path Optimized Link State Protocol with Max-weight Scheduling

Backpressure-based Max-Weight Scheduling (MWS) algorithms are theoretically proven to be optimal with respect to wireless multi-hop networks throughput maximization. On the other hand, Optimized Link State Protocol (OLSR) is the leading proactive protocol for wireless ad-hoc networks. However, in its standard implementation, OLSR is a single-path protocol, whereas it is widely known that wireless network performance can be improved by allowing multiple routes from a given source to a given destination. Moreover, MWS algorithms provide higher throughput when used in multi-path routing scenarios. We designed a light-weight multi-path modification of the OLSR algorithm that is dedicated to work in cooperation with backpressure-based scheduling. The proposed routing algorithm preserves all features of proactive protocols. We tested our joint routing and scheduling scheme in a wireless network serving TCP flows. Since performance of multi-path TCP can suffer from packet reordering, we implemented a delayed reordering algorithm. In order to keep 802.11 MAC unmodified, we deployed a mechanism for the estimation of number of packets in the MAC sub-layer.

Reflective Relational Learning for Ontology Alignment

We propose an application of a statistical relational learning method as a means for automatic detection of semantic correspondences between concepts of OWL ontologies. The presented method is based on an algebraic data representation which, in contrast to well-known graphical models, imposes no arbitrary assumption with regard to the data model structure. We use a probabilistic relevance model as the basis for the estimation of the most plausible matches.We experimentally evaluate the proposed method employing datasets developed for the Ontology Alignment Evaluation Initiative (OAEI) Anatomy track, for the task of identifying matches between concepts of AdultMouse Anatomy ontology and NCI Thesaurus ontology on the basis of expert matches partially provided to the system.

Multi-path OLSR performance analysis in a large testbed environment

Optimized Link State Routing (OLSR) protocol is a leading proactive routing protocol for Mobile Ad-hoc Networks (MANETs). Since the OLSR protocol in its standard version does not support multi-path packet forwarding, we have developed and implemented its multi-path extension. As a result of a slightly modified path computation algorithm and the application of backpressure Max-Weight Scheduling (MWS) policy, the MANETs can utilize the extended functionality based on OLSR. The experiment results presented in this paper compare the performance of the standard and extended OLSR versions in a large congested MANET (i.e., in large-scale DES-Testbed located at Freie Universitat Berlin).

Tensor-Based Modeling of Temporal Features for Big Data CTR Estimation

In this paper we propose a simple tensor-based approach to temporal features modeling that is applicable as means for logistic regression (LR) enhancement. We evaluate experimentally the performance of an LR system based on the proposed model in the Click-Through Rate (CTR) estimation scenario involving processing of very large multi-attribute data streams. We compare our approach to the existing approaches to temporal features modeling from the perspective of the Real-Time Bidding (RTB) CTR estimation scenario. On the basis of an extensive experimental evaluation, we demonstrate that the proposed approach enables achieving an improvement of the quality of CTR estimation. We show this improvement in a Big Data application scenario of the Web user feedback prediction realized within an RTB Demand-Side Platform.

Progressive Random Indexing: Dimensionality Reduction Preserving Local Network Dependencies

The vector space model is undoubtedly among the most popular data representation models used in the processing of large networks. Unfortunately, the vector space model suffers from the so-called curse of dimensionality, a phenomenon where data become extremely sparse due to an exponential growth of the data space volume caused by a large number of dimensions. Thus, dimensionality reduction techniques are necessary to make large networks represented in the vector space model available for analysis and processing. Most dimensionality reduction techniques tend to focus on principal components present in the data, effectively disregarding local relationships that may exist between objects. This behavior is a significant drawback of current dimensionality reduction techniques, because these local relationships are crucial for maintaining high accuracy in many network analysis tasks, such as link prediction or community detection. To rectify the aforementioned drawback, we propose Progressive Random Indexing, a new dimensionality reduction technique. Built upon Reflective Random Indexing, our method significantly reduces the dimensionality of the vector space model while retaining all important local relationships between objects. The key element of the Progressive Random Indexing technique is the use of the gain value at each reflection step, which determines how much information about local relationships should be included in the space of reduced dimensionality. Our experiments indicate that when applied to large real-world networks (Facebook social network, MovieLens movie recommendations), Progressive Random Indexing outperforms state-of-the-art methods in link prediction tasks.

Optimization driven multi-hop network design and experimentation: the approach of the FP7 project OPNEX

The OPNEX project exemplifies system and optimization theory as the foundations for algorithms that provably maximize capacity of wireless networks. The algorithms termed in abstract network models have been converted to protocols and architectures practically applicable to wireless systems. A validation methodology through experimental protocol evaluation in real network testbeds has been proposed and used. OPNEX uses recent advances in system theoretic network control, including the Back-Pressure principle, max-weight scheduling, utility optimization, congestion control, and the primaldual method for extracting network algorithms. These approaches exhibited vast potential for achieving high capacity and full exploitation of resources in abstract network models and found their way to reality in high performance architectures developed as a result of the research conducted within OPNEX.

IMS-based performance analysis of a MANET controlled by the Delay-Aware NUM system

This paper presents the results of the performance evaluation concerning 3G/3GPP IP Multimedia Subsystem (IMS) applications deployed in a wireless multi-hop network controlled by the Delay-Aware Network Utility Maximization System (DANUMS). The main purpose of the DANUM system is simultaneous service of files and multimedia streams, i.e., types of traffic of different delay requirements. DANUMS combines a queue-level-based approximation of Max Weight Scheduling with a multi-path backpressure-oriented extension of the Optimized Link State Routing (OLSR) protocol. In order to evaluate DANUMS in a 3G videoconferencing scenario, we have integrated an Open IMS Core wired network installation with a wireless DANUMS testbed (playing the role of an IMS access network). The results of experiments presented in the paper show that by applying the DANUMS virtual queuing system, we can achieve a utility-aware coexistence of IMS audiovisual streams and non-IMS flows of different delay requirements.