arXiv subjects

Kave Salamatian

Publications and source records attributed to Kave Salamatian.

7 recordsLinked to original sources

arXiv2026

Improving the Robustness/Accuracy Tradeoff Against Adversarial Attacks Using Information Bottleneck Distillation Through Dual Teachers

Deep neural networks (DNNs) have achieved remarkable success in classical machine learning problems. However, they are known to be vulnerable to adversarial attacks. Countermeasures proposed in the literature, notably Information Bottleneck Distillation (IBD) introduced by Kuang et al., degrade the classification accuracy on clean inputs while improving the robustness to adversarial inputs. In this work, we extend the IBD framework by introducing an extra teacher model (clean teacher) trained with only clean inputs, into the distillation process from a robust teacher model trained by adversarial training. The features of both clean and robust teachers are transferred to the student through a cross-layer attention matrix. Experimental results on the CIFAR-10 and CIFAR-100 datasets show that the proposed method improves classification accuracy on clean samples compared to the original IBD, while maintaining similar accuracy on adversarial samples. Furthermore, our methods are competitive with state-of-the-art approaches, including the recent dual-teacher distillation framework B-MTARD, particularly in terms of the harmonic mean between clean and robust accuracy. We also analyze the impact of different training settings that have different influences on the attention module.

cs.LG↗

arXiv2021

Wide-AdGraph: Detecting Ad Trackers with a Wide Dependency Chain Graph

Websites use third-party ads and tracking services to deliver targeted ads and collect information about users that visit them. These services put users' privacy at risk, and that is why users' demand for blocking these services is growing. Most of the blocking solutions rely on crowd-sourced filter lists manually maintained by a large community of users. In this work, we seek to simplify the update of these filter lists by combining different websites through a large-scale graph connecting all resource requests made over a large set of sites. The features of this graph are extracted and used to train a machine learning algorithm with the aim of detecting ads and tracking resources. As our approach combines different information sources, it is more robust toward evasion techniques that use obfuscation or changing the usage patterns. We evaluate our work over the Alexa top-10K websites and find its accuracy to be 96.1% biased and 90.9% unbiased with high precision and recall. It can also block new ads and tracking services, which would necessitate being blocked by further crowd-sourced existing filter lists. Moreover, the approach followed in this paper sheds light on the ecosystem of third-party tracking and advertising.

cs.CR↗

arXiv2018

Pull-based Bloom Filter-based Routing for Information-Centric Networks

In Named Data Networking (NDN), there is a need for routing protocols to populate Forwarding Information Base (FIB) tables so that the Interest messages can be forwarded. To populate FIBs, clients and routers require some routing information. One method to obtain this information is that network nodes exchange routing information by each node advertising the available content objects. Bloom Filter-based Routing approaches like BFR [1], use Bloom Filters (BFs) to advertise all provided content objects, which consumes valuable bandwidth and storage resources. This strategy is inefficient as clients request only a small number of the provided content objects and they do not need the content advertisement information for all provided content objects. In this paper, we propose a novel routing algorithm for NDN called pull-based BFR in which servers only advertise the demanded file names. We compare the performance of pull-based BFR with original BFR and with a flooding-assisted routing protocol. Our experimental evaluations show that pull-based BFR outperforms original BFR in terms of communication overhead needed for content advertisements, average roundtrip delay, memory resources needed for storing content advertisements at clients and routers, and the impact of false positive reports on routing. The comparisons also show that pull-based BFR outperforms flooding-assisted routing in terms of average round-trip delay.

cs.NI↗

arXiv2017

BFR: a Bloom Filter-based Routing Approach for Information-Centric Networks

Locating the demanded content is one of the major challenges in Information-Centric Networking (ICN). This process is known as content discovery. To facilitate content discovery, in this paper we focus on Named Data Networking (NDN) and propose a novel routing scheme for content discovery, called Bloom Filter-based Routing (BFR), which is fully distributed, content oriented, and topology agnostic at the intra-domain level. In BFR, origin servers advertise their content objects using Bloom filters. We compare the performance of the proposed BFR with flooding and shortest path content discovery approaches. BFR outperforms its counterparts in terms of the average round-trip delay, while it is shown to be very robust to false positive reports from Bloom filters. Also, BFR is much more robust than shortest path routing to topology changes. BFR strongly outperforms flooding and performs almost equal with shortest path routing with respect to the normalized communication costs for data retrieval and total communication overhead for forwarding Interests. All the three approaches achieve similar mean hit distance. The signalling overhead for content advertisement in BFR is much lower than the signalling overhead for calculating shortest paths in the shortest path approach. Finally, BFR requires small storage overhead for maintaining content advertisements.

cs.NI↗

arXiv2012

Optimization of Bloom Filter Parameters for Practical Bloom Filter Based Epidemic Forwarding in DTNs

Epidemic forwarding has been proposed as a forwarding technique to achieve opportunistic communication in Delay Tolerant Networks. Even if this technique is well known and widely referred, one has to first deal with several practical problems before using it. In particular, in order to manage the redundancy and to avoid useless transmissions, it has been proposed to ask nodes to exchange information about the buffer content prior to sending information. While Bloom filter has been proposed to transport the buffer content information, up to our knowledge no real evaluation has been provided to study the tradeoff that exists in practice. In this paper we describe an implementation of an epidemic forwarding scheme using Bloom filters. Then we propose some strategies for Bloom filter management based on windowing and describe implementation tradeoffs. By simulating our proposed strategies in ns-3 both with random waypoint mobility and realistic mobility traces coming from San Francisco taxicabs, we show that our proposed strategies alleviate the challenge of using epidemic forwarding in DTNs.

cs.NI↗

arXiv2007

Characterization of P2P IPTV Traffic: Scaling Analysis

P2P IPTV applications arise on the Internet and will be massively used in the future. It is expected that P2P IPTV will contribute to increase the overall Internet traffic. In this context, it is important to measure the impact of P2P IPTV on the networks and to characterize this traffic. Dur- ing the 2006 FIFA World Cup, we performed an extensive measurement campaign. We measured network traffic generated by broadcasting soc- cer games by the most popular P2P IPTV applications, namely PPLive, PPStream, SOPCast and TVAnts. From the collected data, we charac- terized the P2P IPTV traffic structure at different time scales by using wavelet based transform method. To the best of our knowledge, this is the first work, which presents a complete multiscale analysis of the P2P IPTV traffic. Our results show that the scaling properties of the TCP traffic present periodic behavior whereas the UDP traffic is stationary and lead to long- range depedency characteristics. For all the applications, the download traffic has different characteristics than the upload traffic. The signaling traffic has a significant impact on the download traffic but it has negligible impact on the upload. Both sides of the traffic and its granularity has to be taken into account to design accurate P2P IPTV traffic models.

cs.NI↗

arXiv2004

Describing and Simulating Internet Routes

This paper introduces relevant statistics for the description of routes in the internet, seen as a graph at the interface level. Based on the observed properties, we propose and evaluate methods for generating artificial routes suitable for simulation purposes. The work in this paper is based upon a study of over seven million route traces produced by CAIDA's skitter infrastructure.

cs.NI↗