Impressive performance of pre-trained models has garnered public attention and made news headlines in recent years.Almost always, these models are produced by or in collaboration with industry.Using them is critical for competing on natural language processing (NLP) benchmarks and correspondingly to stay relevant in NLP research.We surveyed 100 papers published at EMNLP 2022 to determine the degree to which researchers rely on industry models, other artifacts, and contributions to publish in prestigious NLP venues and found that the ratio of their citation is at least three times greater than what would be expected.Our work serves as a scaffold to enable future researchers to more accurately address whether: 1) Collaboration with industry is still collaboration in the absence of an alternative or 2) if NLP inquiry has been captured by the motivations and research direction of private corporations.

Opacity is a property of a system that captures whether certain event sequences (or certain states) are indistinguishable from other event sequences (or states) in the system. Opacity is used in analyzing privacy, secrecy, and other aspects of systems modeled by discrete-event systems. In this paper, we introduce the concept of minimal information release policies for non-opacity and the concept of mixed opacity. Mixed opacity policies are introduced as a holistic approach for solving problems that involve a combination of releasing information to make some objectives of the system opaque while making some other objectives non-opaque. We present a set of algorithms for information release under a mixed opacity policy. These algorithms compute policies in a system such that two given sublanguages are opaque, and at the same time, two other sublanguages in the same system are non-opaque. The application of mixed opacity is demonstrated on the Dining Cryptographers Problem.

Epidermal Growth Factor Receptor (EGFR) signaling to the Ras-MAPK pathway is implicated in the development and progression of cancer and is a major focus of targeted combination therapies. Physiochemical models have been used for identifying and testing the signal-inhibiting potential of targeted therapies, however, their application to larger multi-pathway networks is limited by the availability of experimentally-determined rate and concentration parameters. An alternate strategy for identifying and evaluating drug-targetable nodes is proposed. A physiochemical model of EGFR-Ras-MAPK signaling is implemented and calibrated to experimental data. Essential topological features of the model are converted into a Petri net and nodes that behave as siphons-a structural property of Petri nets-are identified. Siphons represent potential drug-targets since they are unrecoverable if their values fall below a threshold. Centrality measures are then used to prioritize siphons identified as candidate drug-targets. Single and multiple drug-target combinations are identified which correspond to clinically relevant drug targets and exhibit inhibition synergy in physiochemical simulations of EGF-induced EGFR-Ras-MAPK signaling. Taken together, these studies suggest that siphons and centrality analyses are a promising computational strategy to identify and rank drug-targetable nodes in larger networks as they do not require knowledge of the dynamics of the system, but rely solely on topology.

When designers and developers create software they often overlook issues related to security. Ideally, protection of the program from illegal usage would be considered at each stage of this program's life cycle. The proposition put forward here is to augment intrusion detection systems (IDSs) and employ them as a tool to support secure software development. Many state-based intrusion detection methods share structural and behavioural similarities with the set of processes known as discrete-event systems (DESs). A common structure for modelling DESs is the deterministic finite-state automaton. There exist several compatible anomaly detection techniques which construct finite- state machine models of normal behaviour through the decomposition of associated data (e.g., system calls, HTTP requests) into sequences of events. This paper proposes the application of decentralized DES theory to formally analyze and enhance these approaches to anomaly detection with misuse prevention. Models of misuse attacks are generated in the same manner as the legal usage representation, then augmented and integrated into the program model to prevent the execution of malicious sequences. The technique described herein simultaneously uses anomaly and misuse approaches to prevent and disable attacks before their completion.

The development of controls for the execution of concurrent code is non-trivial. We show how existing discrete-event system (DES) theory can be successfully applied to this problem. From code without concurrency controls and a specification of desired behaviours, concurrency control code is generated. By applying rigorously proven DES theory, we guarantee that the control scheme is nonblocking (and thus free of both deadlock and livelock) and minimally restrictive. Some conflicts between specifications and source can be automatically resolved without introducing new specifications. Moreover, the approach is independent of specific programming or specification languages. Two examples using Java are presented to illustrate the approach. Additional applicable DES results are discussed as future work.