The popular library TENSORFLOW (TF) has familiarised the mainstream of machine-learning community with programming language concepts such as data-flow computing and automatic differentiation. Additionally, it has introduced some genuinely new syntactic and semantic programming concepts. In this paper we study one such new concept, the ability to extract and manipulate the state of a computation graph. This feature allows the convenient specification of parameterised models by freeing the programmer of the bureaucracy of parameter management, while still permitting the use of generic, model-independent, search and optimisation algorithms. We study this new language feature, which we call 'graph abstraction' in the context of the call-by-value lambda calculus, using the recently developed Dynamic Geometry of Interaction formalism. We give a simple type system guaranteeing the safety of graph abstraction, and we also show the safety of critical language properties such as garbage collection and the beta law. The semantic model suggests that the feature could be implemented in a general-purpose functional language reasonably efficiently.

Internet censorship by governments is an increasingly common practice worldwide. Internet users and censors are locked in an arms race: as users find ways to evade censorship schemes, the censors develop countermeasures for the evasion tactics. One of the most popular and effective circumvention tools, Tor, must regularly adjust its network traffic signature to remain usable.

Digital control systems are essential to the safe and efficient operation of a variety of industrial processes in sectors such as electric power, oil and gas, water treatment, and manufacturing. Modern control systems are increasingly connected to other control systems as well as to corporate systems. They are also increasingly adopting networking technology and system and application software from conventional enterprise systems. These trends can make control systems vulnerable to cyber attack, which in the case of control systems may impact physical processes causing environmental harm or injury. We present some results of the DATES (Detection and Analysis of Threats to the Energy Sector) project, wherein we adapted and developed several intrusion detection technologies for control systems. The suite of detection technologies was integrated and connected to a commercial security event correlation framework from ArcSight. We demonstrated the efficacy of our detection and correlation solution on two coupled testbed environments. We particularly focused on detection, correlation, and visualization of a network traversal attack, where an attacker penetrates successive network layers to compromise critical assets that directly control the underlying process. Such an attack is of particular concern in the layered architectures typical of control system implementations.

Firewalls and network intrusion detection systems (NIDSs) are widely used to secure computer networks. Given a network that deploys multiple firewalls and NIDSs, ensuring that these security components are correctly configured is a challenging proble

Many network applications depend on the security of the domain name system (DNS). Attacks on the DNS can cause denial of service and entity authentication to fail. In our approach, we use formal specifications to characterize DNS clients and DNS name servers, and to define a security goal. A name server should only use DNS data that is consistent with data from name servers that manage the corresponding domains (i.e. authoritative name servers). To enforce the security goal, we formally specify a DNS wrapper that examines the incoming and outgoing DNS messages of a name server to detect messages that could cause violations of the security goal, cooperates with the corresponding authoritative name servers to diagnose those messages, and drops the messages that are identified as threats. Based on the wrapper specification, we implemented a wrapper prototype and evaluated its performance. Out experiments show that the wrapper incurs reasonable overhead and is effective against DNS attacks such as cache poisoning and certain spoofing attacks.

An attractive target for a computer system attacker is the router. An attacker in control of a router can disrupt communication by dropping or misrouting packets passing through the router. We present a protocol called WATCHERS that detects and reacts to routers that drop or misroute packets. WATCHERS is based on the principle of conservation of flow in a network: all data bytes sent into a node, and not destined for that node, are expected to exit the node. WATCHERS tracks this flow, and detects routers that violate the conservation principle. We show that WATCHERS has several advantages over existing network monitoring techniques. We argue that WATCHERS' impact on router performance and WATCHERS' memory requirements are reasonable for many environments. We demonstrate that in ideal conditions WATCHERS makes no false-positive diagnoses. We also describe how WATCHERS can be tuned to perform nearly as well in realistic conditions.

We present a solution to the denial of service problem for routing infrastructures. When a network suffers from denial of service, packets cannot reach their destinations. Existing routing protocols are not well-equipped to deal with denial of service; a misbehaving router---which may be caused by software/hardware faults, misconfiguration, or malicious attacks---may be able to disable entire networks. To protect network infrastructures from routers that incorrectly drop packets and misroute packets, we hypothesize failure models for routers and present protocols that detect and respond to those misbehaving routers. Based on realistic assumptions, we prove that our protocols have the following properties: (1) A well-behaved router never incorrectly claims another router as a misbehaving router; (2) If a network has misbehaving routers, one or more of them can be located; (3) Misbehaving routers will eventually be removed. 1 Introduction Through a myriad of applications, including elec...