This paper investigates the rational thinking capability of Large Language Models (LLMs) in multi-round argumentative debates by exploring the impact of fallacious arguments on their logical reasoning performance. More specifically, we present Logic Competence Measurement Benchmark (LOGICOM), a diagnostic benchmark to assess the robustness of LLMs against logical fallacies. LOGICOM involves two agents: a persuader and a debater engaging in a multi-round debate on a controversial topic, where the persuader tries to convince the debater of the correctness of its claim. First, LOGICOM assesses the potential of LLMs to change their opinions through reasoning. Then, it evaluates the debater’s performance in logical reasoning by contrasting the scenario where the persuader employs logical fallacies against one where logical reasoning is used. We use this benchmark to evaluate the performance of GPT-3.5 and GPT-4 using a dataset containing controversial topics, claims, and reasons supporting them. Our findings indicate that both GPT-3.5 and GPT-4 can adjust their opinion through reasoning. However, when presented with logical fallacies, GPT-3.5 and GPT-4 are erroneously convinced 41% and 69% more often, respectively, compared to when logical reasoning is used. Finally, we introduce a new dataset containing over 5k pairs of logical vs. fallacious arguments.

When searching a database for a topic (e.g. Covid-19), there may not exist a precise match, especially if the topic is novel. Furthermore, the topic may surface in the data under different guises (`Covid-19,' `coronavirus,' `pandemic', etc.). The results of a keyword search are limited by the querier's imagination and familiarity with the data. Such searches have high precision, but low recall. In order to increase the recall of searches, we present the Semantic Search Pipeline, a novel approach to document retrieval that uses distributional semantic models and locality sensitive hashing to expand queries and efficiently identify other relevant documents that may not contain the obvious query terms. We evaluate the pipeline using a dataset curated from financial customer service call centers, resulting in an increase in recall of 32% over a simple keyword baseline, with a negligible drop in precision. Furthermore, we present the notion of concept forging, a process of tracing a topic or concept through time and through its various surface realizations. Applied to Covid-19, the search pipeline retrieves a set of documents that allow us to uncover the short- and long-term effects of Covid-19 on the lives of the people and businesses impacted by it. Although Covid-19 is a timely test case, our search pipeline is general in nature and can be easily applied to any range of topics.

The Domain Name System (DNS), a fundamental protocol that controls how users interact with the Internet, inadequately provides protection for user privacy. Recently, there have been advancements in the field of DNS privacy and security in the form of the DNS over TLS (DoT) and DNS over HTTPS (DoH) protocols. The advent of these protocols and recent advancements in large-scale data processing have drastically altered the threat model for DNS privacy. Users can no longer rely on traditional methods, and must instead take active steps to ensure their privacy. In this paper, we demonstrate how the extended Berkeley Packet Filter (eBPF) can assist users in maintaining their privacy by leveraging eBPF to provide privacy across standard DNS, DoH, and DoT communications. Further, we develop a method that allows users to enforce application-specific DNS servers. Our method provides users with control over their DNS network traffic and privacy without requiring changes to their applications while adding low overhead.

Ambiguity is inherent to human language and poses a unique challenge both to human listeners as well as natural language processing (NLP) systems. Ambiguity is understood as a type of uncertainty which allows for more than one plausible interpretation of utterances. Ambiguity can introduce problems for NLP systems designed to, for example, execute machine translation, determine sentiment, and perform automatic speech recognition (ASR). We seek to identify and resolve mis-transcriptions that arise from phonetic ambiguity and degraded acoustic signals in 87,000 call transcripts. We first present an alignment algorithm which identifies mis-transcriptions generated by ASR systems when compared against verified human transcriptions. This method not only allows for a general evaluation of ASR performance but also highlights specific areas of difficulty for such systems (e.g. “considerate” vs. “consider it”). We further present an error resolution algorithm which, given a mis-transcribed word, uses contextual cues to suggest a more likely, phonetically similar word. This work has the potential to not only evaluate existing ASR systems, but also to immediately improve their performance.

The number of IoT devices is expected to be between 32-99 Billion by 2025, many of which will use the cellular wireless data network for communications. This presents a unique challenge to the operator while allocating resources, namely how to optimally balance CPU and memory usage in virtualized and physical hosts while simultaneously handling millions of IoT devices without affecting the quality of experience of normal mobile users. Due to the sheer number of the IoT devices, it is not feasible to store their session context in memory. In this work, we present a machine learning model that predicts the network usage pattern of five broad classes of cIoT devices. The prediction model trained on a Multilayer Perceptron allows the network operator to opportunistically prefetch cIoT context from secondary storage before it is required. Further, we propose a new metric - Value of Perfect Information - to assess our approach. We evaluate our approach across two fronts: First, we study the efficacy of replacement algorithms such as LRU, MRU, FIFO and random replacement; we also assess the impact of varying memory slots. Finally, we evaluate our models against the default (no prefetching) model and an on-time prefetching model to demonstrate the value of our pre-fetching approach.

Modern cellular networks are complex systems offering a wide range of services and present challenges in detecting anomalous events when they do occur. The networks are engineered for high reliability and, hence, the data from these networks is predominantly normal with a small proportion being anomalous. From an operations perspective, it is important to detect these anomalies in a timely manner, to correct vulnerabilities in the network and preclude the occurrence of major failure events. The objective of our work is anomaly detection in cellular networks in near real-time to improve network performance and reliability. We use performance data from a 4G LTE network to develop a methodology for anomaly detection in such networks. Two rigorous prediction models are proposed: a non-parametric approach (Chi-Square test), and a parametric one (Gaussian Mixture Models). These models are trained to detect differences between distributions to classify a target distribution as belonging to a normal period or abnormal period with high accuracy. We discuss the merits between the approaches and show that both provide a more nuanced view of the network than simple thresh-olds of success/failure used by operators in production networks today.

The wireless/cellular communications network is composed of a complex set of interconnected computation units that form the mobile core network. The mobile core network is engineered to be fault tolerant and redundant; small errors that manifest themselves in the network are usually resolved automatically. However, some errors remain latent, and if discovered early enough can provide warnings to the network operator about a pending service outage. For mobile network operators, it is of high interest to detect these minor anomalies near real-time. In this work we use performance data from a 4G-LTE network carrier to train two parameter-free models. A first model relies on isolation forests, and the second is histogram based. The trained models represent the data characteristics for normal periods; new data is matched against the trained models to classify the new time period as being normal or abnormal. We show that the proposed methods can gauge the mobile network state with more subtlety than standard success/failure thresholds used in real-world networks today.

Due to their promise of delivering real-time network insights, today's streaming analytics platforms are increasingly being used in the communications networks where the impact of the insights go beyond sentiment and trend analysis to include real-time detection of security attacks and prediction of network state (i.e., is the network transitioning towards an outage). Current streaming analytics platforms operate under the assumption that arriving traffic is to the order of kilobytes produced at very high frequencies. However, communications networks, especially the telecommunication networks, challenge this assumption because some of the arriving traffic in these networks is to the order of gigabytes, but produced at medium to low velocities. Furthermore, these large datasets may need to be ingested in their entirety to render network insights in real-time. Our interest is to subject today's streaming analytics platforms --- constructed from state-of-the art software components (Kafka, Spark, HDFS, ElasticSearch) --- to traffic densities observed in such communications networks. We find that filtering on such large datasets is best done in a common upstream point instead of being pushed to, and repeated, in downstream components. To demonstrate the advantages of such an approach, we modify Apache Kafka to perform limited native data transformation and filtering, relieving the downstream Spark application from doing this. Our approach outperforms four prevalent analytics pipeline architectures with negligible overhead compared to standard Kafka. (Our modifications to Apache Kafka are publicly available at https://github.com/Esquive/queryable-kafka.git)

The next-generation 5G mobile network architecture will support the rapid deployment of new, dynamic network services that are capable of responding to current network conditions and demands. Software-defined Networking (SDN), virtualization technologies, and real-time analytics are the core components that will enable an adaptive and responsive 5G network. We present a case study of a real-time communications (RTC) video service that highlights the manner in which the core components (SDN, virtualization, analytics) allow a flexible and elastic 5G network. Because an end-to-end 5G network does not exist today, we construct one using artifacts from the current 4G/LTE network to host our dynamic network enabled RTC service. We identify three main insights from executing our service that could prove beneficial to the 5G network evolution: need for efficient horizontal control, need to limit identifier proliferation, and the existence of control-plane network functions in service network-function graphs.

The U.S. National Academies of Science's Board on Science, Technology and Economic Policy estimates that the Internet and voice-over-IP (VoIP) communications infrastructure generates 10% of U.S. economic growth. As market forces move increasingly towards Internet and VoIP communications, there is proportional increase in telephony denial of service (TDoS) attacks. Like denial of service (DoS) attacks, TDoS attacks seek to disrupt business and commerce by directing a flood of anomalous traffic towards key communication servers. In this work, we focus on a new class of anomalous traffic that exhibits a mimicry TDoS attack. Such an attack can be launched by crafting malformed messages with small changes from normal ones. We show that such malicious messages easily bypass intrusion detection systems (IDS) and degrade the goodput of the server drastically by forcing it to parse the message looking for the needed token. Our approach is not to parse at all; instead, we use multiple classifier systems (MCS) to exploit the strength of multiple learners to predict the true class of a message with high probability (98.50% ≤ p ≤ 99.12%). We proceed systematically by first formulating an optimization problem of picking the minimum number of classifiers such that their combination yields the optimal classification performance. Next, we analytically bound the maximum performance of such a system and empirically demonstrate that it is possible to attain close to the maximum theoretical performance across varied datasets. Finally, guided by our analysis we construct an MCS appliance that demonstrates superior classification accuracy with O(1) runtime complexity across varied datasets.

The Software Defined Network (SDN) model depends on abstractions to separate the control plane from the packet forwarding plane. Applications can interact with the control plane to receive a global network view, upon which they can operate. By having access to network topology information, applications can optimize decisions related to service rendezvous, service fulfillment, service placement and service removal. The network is in the best position to provide guidance to a broad class of applications, including peer-to-peer systems, Content Distribution Network (CDN), and datacenter applications. In all these use cases proximity matters as peers need to rendezvous with other peers and users need to rendezvous with the best cloud application or best CDN server. We maintain that a solution for such a rendezvous problem should be an intrinsic component of the emerging SDN model. A specific instance of a protocol that abstracts network topology is the Application Layer Traffic Optimization (ALTO) protocol. ALTO provides applications an abstract view of the network and thus enables applications to leverage a network without exposing the network provider's internal details or policies. We argue that ALTO provides a clean, mature, standards-based and powerful abstraction, which can be used by SDNs today to obtain network information for solving the rendezvous problem.

The Session Initiation Protocol (SIP) is an important multimedia session establishment protocol used on the Internet. It is a text-based protocol, which is complex to parse due to the wide variability in representing the information elements. Building a parser for SIP may appear straight-forward; however, writing an efficient, robust, and scalable parser that is immune to low-effort attacks using malformed messages is surprisingly difficult. To mitigate this, self-learning systems based on Euclidean distance classifiers have been proposed to determine whether a message is well-formed or not. The efficacy of such machine learning algorithms must be studied on varied data sets before they can be successfully used. Our previous work has shown that Euclidean distance-based classifiers and standard classifiers used for self-learning problems are unable to detect malformed self-similar SIP messages (i.e., invalid SIP messages that differ by only a few bytes from normal SIP messages). This paper proposes using multiple classifier systems to detect malformed self-similar SIP messages. Our results show that a judiciously constructed multiple classifier system yields classification performance as high as 97.56% of the messages being classified correctly. We further show that for self-similar SIP messages, feature reduction measures based on the first moment are insufficient for improving classification accuracy.

We consider a system of compute and storage resources geographically distributed over a large number of locations connected via a wide-area network. By distributing the resources, latency to users can be decreased, bandwidth costs reduced and availablility increased. The challenge is to distribute services with varying characteristics among the data centers optimally. Some services are very latency sensitive, others need vast amounts of storage, and yet others are computationally complex but do not require hard deadlines on execution. We propose efficient algorithms for the placement of services to get the maximum benefit from a distributed cloud systems. The algorithms need input on the status of the network, compute resources and data resources, which are matched to application requirements.

The Session Initiation Protocol (SIP) is an important multimedia session establishment protocol used on the Internet. Due to the nature and deployment realities of the protocol (ASCII message representation, most deployments over UDP, limited use of message encryption), it becomes relatively easy to attack the protocol at the message level. To mitigate this, self-learning systems have been proposed to counteract new threats. However the efficacy of existing machine learning algorithms must be studied on varied data sets before they can be successfully used. Existing literature indicates that Euclidean distance based classifiers work well to detect anomalous messages. Our work suggests that such classifiers do not produce adequate results for well-crafted malicious messages that differ very slightly from normal messages. To demonstrate this, we gather SIP traffic and minimally perturb it using 13 generic transforms to create malicious SIP messages. We use the Levenshtein distance, L, as a measure of similarity between normal and malicious SIP messages. We subject our dataset — consisting of malicious and normal SIP messages — to Euclidean distance-based classifiers as well as four standard classifiers. Our results show vast differences for Euclidean distance-based classifiers on our dataset than reported in current literature. We further see that the standard classifiers are better able to classify an anomalous message when L is small.

Abstract We definephishingas the practice of directing unsuspecting users to fraudulent websites with the intent of obtaining personal information to be used for illicit purpose by a spammer. We introduce a new anti‐phishing filter, phishwish, that has a number of advantages over existing phishing filters: it does not need to be trained, as is the case with Bayesian filters, nor does it consult centralized white or blacklists to determine whether an email is suspect. Phishwish uses a set of only 11 rules to determine the veracity of an incoming email; the results can be used to quarantine the email or to alert the user. We compare the performance of phishwish to SpamAssassin, a popular open source filter, as well as the Google phishing filters accessed from the Firefox browser. Our results indicate that phishwish outperforms existing filters in identifying phishing emails, even identifying those originated by the ‘rock phish’ gang, and that it aids in detection of zero‐day attacks that were not caught by existing filters. Copyright © 2008 John Wiley & Sons, Ltd.

Proxies provide important rendezvous service in the session initiation protocol (SIP), but it comes at a cost to privacy. A SIP proxy is privy to all of the signaling exchanged between two user agents, even if that signaling is performed over a secure channel (e.g., a Transport Layer Security channel.) This paper proposes and evaluates a mechanism that allows the proxies to create an overlay network between the user agents for rendezvous, and once that is done, the proxies become transparent traffic forwarders. From then onwards, user agents can authenticate each other directly and exchange cryptographically secure signaling traffic over the overlay network created by the proxies. This mechanism is applicable to traditional client/server SIP as well as Peer-to-Peer SIP.

Open source practices and tools have proven to be highly effective for overcoming the many problems of geographically distributed software development. We know relatively little, however, about the range of settings in which they work. In particular, can corporations use the open source development model effectively for software projects inside the corporate domain? Or are these tools and practices incompatible with development environments, management practices, and market-driven schedule and feature decisions typical of a commercial software house? We present a case study of open source software development methodology adopted by a significant commercial software project in the telecommunications domain. We extract a number of lessons learned from the experience, and identify open research questions.

We discuss extensions to the session initiation protocol (SIP) to enable a rich set of services between the cellular public switched telephone network (PSTN) and the Internet. We consider all manner of cellular endpoints, 2G, 2.5G, and early 3G endpoints. These extensions allow the PSTN endpoints to act as smart SIP user agents and participate in Internet-oriented services, such as presence and instant messaging (IM), without the endpoint itself being connected to the Internet. For cellular networks, our methodology conserves precious radio spectrum by offloading such data services from the bandwidth normally allocated to voice. Our approach is based on open standards and is not an ad-hoc or incremental solution.

We propose a general purpose service architecture for realizing services which start in the Public Switched Telephone Network (PSTN) but terminate and execute on the Internet. We discuss the needs for such services, our early research efforts in this direction which lead to prototyping certain benchmark services, and the current state of work in this area. We demonstrate the feasibility of the architecture by focusing on services which involve wireline PSTN as well as the wireless aspects (2 G, 2.5 G) of the PSTN. Our methodology is attractive since it keeps each of the domains (PSTN and Internet) unaware as to where the service is executing with respect to which domain actually requested the service. Individual entities participating in the service do not have any knowledge that external entities from another domain also contributed in the execution and fulfillment of such services. Our approach, as embodied in the service architecture, is to leverage the best of the Internet protocols (SIP, XML, HTTP) and technologies (instant messaging, presence) to provide a general framework for personalized service specification and execution.

Computer networks exist to provide services to users. In domains where more than one network dominates and provides useful services, users will want to avail themselves of services on either of the networks. This paper looks at the convergence of two networks in the telecommunication domain: the Internet and the public switched telephone network and discusses techniques to access services between the two networks, including an architecture for realizing services which would not be possible if either of the networks was operating in isolation.

Networks exist to provide services to users. Increasingly, the networks on which the services reside are not the same as the networks from which the services are accessed. This leads to the problem on how to best provide such services transparently when the access protocols differ. We discusses a methodology to make available the existing services residing in a network whose protocols are distinct from the network where the access attempt is made. We propose a technique we term call model mapping with state sharing and demonstrate it in the telecommunication domain where we access traditional telephony services residing on the telephone network from Internet telephony endpoints residing on the Internet.