Rong Su

Yuting Zhu, Liyong Lin, Rong Su

In this paper, we propose and address the problem of supervisor obfuscation against actuator enablement attack, in a common setting where the actuator attacker can eavesdrop the control commands issued by the supervisor. We propose a method to obfuscate an (insecure) supervisor to make it resilient against actuator enablement attack in such a way that the behavior of the original closed-loop system is preserved. An additional feature of the obfuscated supervisor, if it exists, is that it has exactly the minimum number of states among the set of all the resilient and behavior-preserving supervisors. Our approach involves a simple combination of two basic ideas: 1) a formulation of the problem of computing behavior-preserving supervisors as the problem of computing separating finite state automata under controllability and observability constraints, which can be efficiently tackled by using modern SAT solvers, and 2) the use of a recently proposed technique for the verification of attackability in our setting, with a normality assumption imposed on both the actuator attackers and supervisors.

Liyong Lin, Rong Su

In this paper, we investigate the problem of synthesizing resilient supervisors against combined actuator and sensor attacks, for the subclass of cyber-physical systems that can be modelled as discrete-event systems. We assume that the attackers can carry out actuator enablement and disablement attacks as well as sensor replacement attacks. We consider both risky attackers and covert attackers in the setup where the (partial-observation) attackers may or may not eavesdrop the control commands (issued by the supervisor). A constraint-based approach for the bounded synthesis of resilient supervisors is developed, by reducing the problem to the Quantified Boolean Formulas (QBF) problem. The bounded synthesis problem can then be solved either with a QBF solver or with repeated calls to a propositional satisfiability (SAT) solver, by employing maximally permissive attackers, which can be synthesized with the existing partial-observation supervisor synthesis procedures, as counter examples in the counter example guided inductive synthesis loop.

Liyong Lin, Tomáš Masopust, W. Murray Wonham et al.

A reduction of a source distribution is a collection of smaller sized distributions that are collectively equivalent to the source distribution with respect to the property of decomposability. That is, an arbitrary language is decomposable with respect to the source distribution if and only if it is decomposable with respect to each smaller sized distribution (in the reduction). The notion of reduction of distributions has previously been proposed to improve the complexity of decomposability verification. In this work, we address the problem of generating (optimal) reductions of distributions automatically. A (partial) solution to this problem is provided, which consists of 1) an incremental algorithm for the production of candidate reductions and 2) a reduction validation procedure. In the incremental production stage, backtracking is applied whenever a candidate reduction that cannot be validated is produced. A strengthened substitution-based proof technique is used for reduction validation, while a fixed template of candidate counter examples is used for reduction refutation; put together, they constitute our (partial) solution to the reduction verification problem. In addition, we show that a recursive approach for the generation of (small) reductions is easily supported.

Liyong Lin, Sander Thuijsman, Yuting Zhu et al.

In this work, we propose and develop a new discrete-event based actuator attack model on the closed-loop system formed by the plant and the supervisor. We assume the actuator attacker partially observes the execution of the closed-loop system and eavesdrops the control commands issued by the supervisor. The attacker can modify each control command on a specified subset of attackable events. The attack principle of the actuator attacker is to remain covert until it can establish a successful attack and lead the attacked closed-loop system into generating certain damaging strings. We present a characterization for the existence of a successful attacker, via a new notion of attackability, and prove the existence of the supremal successful actuator attacker, when both the supervisor and the attacker are normal (that is, unobservable events to the supervisor cannot be disabled by the supervisor and unobservable events to the attacker cannot be attacked by the attacker). Finally, we present an algorithm to synthesize the supremal successful attackers that are represented by Moore automata.

Ruikang Luo, Yaofeng Song, Han Zhao et al.

Accurate vehicle type classification serves a significant role in the intelligent transportation system. It is critical for ruler to understand the road conditions and usually contributive for the traffic light control system to response correspondingly to alleviate traffic congestion. New technologies and comprehensive data sources, such as aerial photos and remote sensing data, provide richer and high-dimensional information. Also, due to the rapid development of deep neural network technology, image based vehicle classification methods can better extract underlying objective features when processing data. Recently, several deep learning models have been proposed to solve the problem. However, traditional pure convolutional based approaches have constraints on global information extraction, and the complex environment, such as bad weather, seriously limits the recognition capability. To improve the vehicle type classification capability under complex environment, this study proposes a novel Densely Connected Convolutional Transformer in Transformer Neural Network (Dense-TNT) framework for the vehicle type classification by stacking Densely Connected Convolutional Network (DenseNet) and Transformer in Transformer (TNT) layers. Three-region vehicle data and four different weather conditions are deployed for recognition capability evaluation. Experimental findings validate the recognition ability of our proposed vehicle classification model with little decay, even under the heavy foggy weather condition.

Liyong Lin, Yuting Zhu, Rong Su

In this paper, we shall formulate and address a problem of covert actuator attacker synthesis for cyber-physical systems that are modelled by discrete-event systems. We assume the actuator attacker partially observes the execution of the closed-loop system and is able to modify each control command issued by the supervisor on a specified attackable subset of controllable events. We provide straightforward but in general exponential-time reductions, due to the use of subset construction procedure, from the covert actuator attacker synthesis problems to the Ramadge-Wonham supervisor synthesis problems. It then follows that it is possible to use the many techniques and tools already developed for solving the supervisor synthesis problem to solve the covert actuator attacker synthesis problem for free. In particular, we show that, if the attacker cannot attack unobservable events to the supervisor, then the reductions can be carried out in polynomial time. We also provide a brief discussion on some other conditions under which the exponential blowup in state size can be avoided. Finally, we show how the reduction based synthesis procedure can be extended for the synthesis of successful covert actuator attackers that also eavesdrop the control commands issued by the supervisor.

Ahmad Reza Shehabinia, Liyong Lin, Rong Su

In this paper, we model an operational planning and scheduling problem under multiple job deadlines in a time-weighted automaton framework. We first present a method to determine whether all given job specifications and deadlines can be met by computing a supremal controllable job satisfaction sublanguage. When this supremal sublanguage is not empty, we compute one of its controllable sublanguages that ensures the minimum total job earliness by adding proper delays. When this supremal sublanguage is empty, we will determine the minimal sets of job deadlines that need to be relaxed.

Ruikang Luo, Yaofeng Song, Liping Huang et al.

Accurate long series forecasting of traffic information is critical for the development of intelligent traffic systems. We may benefit from the rapid growth of neural network analysis technology to better understand the underlying functioning patterns of traffic networks as a result of this progress. Due to the fact that traffic data and facility utilization circumstances are sequentially dependent on past and present situations, several related neural network techniques based on temporal dependency extraction models have been developed to solve the problem. The complicated topological road structure, on the other hand, amplifies the effect of spatial interdependence, which cannot be captured by pure temporal extraction approaches. Additionally, the typical Deep Recurrent Neural Network (RNN) topology has a constraint on global information extraction, which is required for comprehensive long-term prediction. This study proposes a new spatial-temporal neural network architecture, called Spatial-Temporal Graph-Informer (STGIN), to handle the long-term traffic parameters forecasting issue by merging the Informer and Graph Attention Network (GAT) layers for spatial and temporal relationships extraction. The attention mechanism potentially guarantees long-term prediction performance without significant information loss from distant inputs. On two real-world traffic datasets with varying horizons, experimental findings validate the long sequence prediction abilities, and further interpretation is provided.

Gang Chen, Yu Lu, Rong Su et al.

Machine learning-based methods have achieved successful applications in machinery fault diagnosis. However, the main limitation that exists for these methods is that they operate as a black box and are generally not interpretable. This paper proposes a novel neural network structure, called temporal logic neural network (TLNN), in which the neurons of the network are logic propositions. More importantly, the network can be described and interpreted as a weighted signal temporal logic. TLNN not only keeps the nice properties of traditional neuron networks but also provides a formal interpretation of itself with formal language. Experiments with real datasets show the proposed neural network can obtain highly accurate fault diagnosis results with good computation efficiency. Additionally, the embedded formal language of the neuron network can provide explanations about the decision process, thus achieve interpretable fault diagnosis.

Ruikang Luo, Yaofeng Song, Liping Huang et al.

Electric Vehicle (EV) charging demand and charging station availability forecasting is one of the challenges in the intelligent transportation system. With the accurate EV station situation prediction, suitable charging behaviors could be scheduled in advance to relieve range anxiety. Many existing deep learning methods are proposed to address this issue, however, due to the complex road network structure and comprehensive external factors, such as point of interests (POIs) and weather effects, many commonly used algorithms could just extract the historical usage information without considering comprehensive influence of external factors. To enhance the prediction accuracy and interpretability, the Attribute-Augmented Spatial-Temporal Graph Informer (AST-GIN) structure is proposed in this study by combining the Graph Convolutional Network (GCN) layer and the Informer layer to extract both external and internal spatial-temporal dependence of relevant transportation data. And the external factors are modeled as dynamic attributes by the attribute-augmented encoder for training. AST-GIN model is tested on the data collected in Dundee City and experimental results show the effectiveness of our model considering external factors influence over various horizon settings compared with other baselines.

Wei Guo, Rong Su, Renhao Tan et al.

CTR prediction, which aims to estimate the probability that a user will click an item, plays a crucial role in online advertising and recommender system. Feature interaction modeling based and user interest mining based methods are the two kinds of most popular techniques that have been extensively explored for many years and have made great progress for CTR prediction. However, (1) feature interaction based methods which rely heavily on the co-occurrence of different features, may suffer from the feature sparsity problem (i.e., many features appear few times); (2) user interest mining based methods which need rich user behaviors to obtain user's diverse interests, are easy to encounter the behavior sparsity problem (i.e., many users have very short behavior sequences). To solve these problems, we propose a novel module named Dual Graph enhanced Embedding, which is compatible with various CTR prediction models to alleviate these two problems. We further propose a Dual Graph enhanced Embedding Neural Network (DG-ENN) for CTR prediction. Dual Graph enhanced Embedding exploits the strengths of graph representation with two carefully designed learning strategies (divide-and-conquer, curriculum-learning-inspired organized learning) to refine the embedding. We conduct comprehensive experiments on three real-world industrial datasets. The experimental results show that our proposed DG-ENN significantly outperforms state-of-the-art CTR prediction models. Moreover, when applying to state-of-the-art CTR prediction models, Dual graph enhanced embedding always obtains better performance. Further case studies prove that our proposed dual graph enhanced embedding could alleviate the feature sparsity and behavior sparsity problems. Our framework will be open-source based on MindSpore in the near future.

Chen Song, Vladimir Cvetkovic, Angela Fontan et al.

The theory of planned behavior (TPB) is one of the most influential frameworks in social psychology, stating that a person's behavior is driven by intention, which is primarily shaped by attitude, subjective norms, and perceived behavioral control. Despite its strong empirical support, TPB remains a static conceptual framework without explicit mathematical formulations that capture the temporal evolution of its components. To address this gap, we develop a dynamic agent-based modeling framework that integrates the core principles of TPB with a behavior-to-attitude feedback mechanism. Specifically, we define behaviors based on their feedback effects on attitude and examine when the population undergoes collective transitions by either adopting a beneficial behavior or rejecting a harmful one. Results from our model demonstrate that collective transitions can be effectively controlled by adjusting two key behavioral parameters that reflect agents' attitude influence and decision rationality. These findings provide quantitative insights on TPB, highlighting the key factors that drive collective behavioral transitions and the need for further socio-psychological case studies.

Chen Song, Angela Fontan, Rong Su et al.

This paper presents a theoretical convergence analysis for an opinion-action coevolution model that integrates the opinion updating rule of the Hegselmann-Krause model with a utility-based decision-making mechanism. The model is reformulated into an augmented state-space representation, where the state matrix induces a time-varying social interaction digraph. The convergence analysis is grounded on two existing theoretical findings that establish convergence for the Hegselmann-Krause type of models and containment control systems with multiple stationary leaders, respectively. Results indicate that, if the structure of the interaction digraph stabilizes within finite time, the model either converges to consensus, where all agents' opinions and actions reach an identical state, or exhibits clustering, where some opinion nodes act as stationary leaders while the remaining nodes approach the convex hull formed by the leaders. Numerical simulations are then provided to validate the theoretical results.

Avalpreet Singh Brar, Rong Su, Jaskaranveer Kaur et al.

Human driver participation is a critical source of uncertainty in Mobility-on-Demand (MoD) rebalancing. Drivers follow platform recommendations probabilistically, and their willingness to comply evolves with experienced outcomes. This creates a closed-loop feedback in which stronger recommendations increase participation, participation increases congestion, congestion lowers allocation success, and realized allocations update adherence beliefs. We propose a microscopic stochastic model that couples (i) belief-driven participation, (ii) Poisson demand, (iii) uniform matching, and (iv) Beta--Bernoulli belief updates. Under a large-population closure, we derive a deterministic mean-field recursion for the population adherence state under platform actuation. For i.i.d. Poisson demand and constant recommendation intensity, we prove global well-posedness and invariance of the recursion, establish equilibrium existence, provide uniqueness conditions, and show global convergence in the regime where platform recommendations are no weaker than baseline participation. We then define steady-state adherence and throughput, characterize the induced performance frontier, and show that adherence and throughput cannot, in general, be simultaneously maximized under uniform time-invariant actuation. This yields a throughput-maximization problem with an adherence floor. Exploiting the monotone frontier structure, we show the optimal uniform time-invariant policy is the maximal feasible recommendation intensity and provide an efficient bisection-based algorithm.

Jiarui Qin, Weinan Zhang, Rong Su et al.

Prediction over tabular data is an essential task in many data science applications such as recommender systems, online advertising, medical treatment, etc. Tabular data is structured into rows and columns, with each row as a data sample and each column as a feature attribute. Both the columns and rows of the tabular data carry useful patterns that could improve the model prediction performance. However, most existing models focus on the cross-column patterns yet overlook the cross-row patterns as they deal with single samples independently. In this work, we propose a general learning framework named Retrieval & Interaction Machine (RIM) that fully exploits both cross-row and cross-column patterns among tabular data. Specifically, RIM first leverages search engine techniques to efficiently retrieve useful rows of the table to assist the label prediction of the target row, then uses feature interaction networks to capture the cross-column patterns among the target row and the retrieved rows so as to make the final label prediction. We conduct extensive experiments on 11 datasets of three important tasks, i.e., CTR prediction (classification), top-n recommendation (ranking) and rating prediction (regression). Experimental results show that RIM achieves significant improvements over the state-of-the-art and various baselines, demonstrating the superiority and efficacy of RIM.

Xiangli Yang, Qing Liu, Rong Su et al.

Recommender systems are often asked to serve multiple recommendation scenarios or domains. Fine-tuning a pre-trained CTR model from source domains and adapting it to a target domain allows knowledge transferring. However, optimizing all the parameters of the pre-trained network may result in over-fitting if the target dataset is small and the number of parameters is large. This leads us to think of directly reusing parameters in the pre-trained model which represent more general features learned from multiple domains. However, the design of freezing or fine-tuning layers of parameters requires much manual effort since the decision highly depends on the pre-trained model and target instances. In this work, we propose an end-to-end transfer learning framework, called Automatic Fine-Tuning (AutoFT), for CTR prediction. AutoFT consists of a field-wise transfer policy and a layer-wise transfer policy. The field-wise transfer policy decides how the pre-trained embedding representations are frozen or fine-tuned based on the given instance from the target domain. The layer-wise transfer policy decides how the high?order feature representations are transferred layer by layer. Extensive experiments on two public benchmark datasets and one private industrial dataset demonstrate that AutoFT can significantly improve the performance of CTR prediction compared with state-of-the-art transferring approaches.

Rong Su

Cybersecurity of discrete event systems (DES) has been gaining more and more attention recently, due to its high relevance to the so-called 4th industrial revolution that heavily relies on data communication among networked systems. One key challenge is how to ensure system resilience to sensor and/or actuator attacks, which may tamper data integrity and service availability. In this paper we focus on some key decidability issues related to smart sensor attacks. We first present a sufficient and necessary condition that ensures the existence of a smart sensor attack, which reveals a novel demand-supply relationship between an attacker and a controlled plant, represented as a set of risky pairs. Each risky pair consists of a damage string desired by the attacker and an observable sequence feasible in the supervisor such that the latter induces a sequence of control patterns, which allows the damage string to happen. It turns out that each risky pair can induce a smart weak sensor attack. Next, we show that, when the plant, supervisor and damage language are regular, it is computationally feasible to remove all such risky pairs from the plant behaviour, via a genuine encoding scheme, upon which we are able to establish our key result that the existence of a nonblocking supervisor resilient to smart sensor attacks is decidable. To the best of our knowledge, this is the first result of its kind in the DES literature on cyber attacks. The proposed decision process renders a specific synthesis procedure that guarantees to compute a resilient supervisor whenever it exists, which so far has not been achieved in the literature.

Rong Su

One of the major challenges about cyber physical systems is how to prevent cyber attacks to ensure system integrity. There has been a large number of different types of attacks discussed in the modern control and computer science communities. In this paper we aim to investigate one special type of attacks in the discrete-event system framework, where an attacker can arbitrarily alter sensor readings after intercepting them from a target system in order to trick a given supervisor to issue control commands improperly, driving the system to an undesirable state. We first consider the cyber attack problem from an attacker point of view, and formulate an attack with bounded sensor reading alterations (ABSRA) problem. We then show that the supremal (or least restrictive) ABSRA exists and can be synthesized, as long as the plant model and the supervisor model are regular, i.e., representable by finite-state automata. Upon the synthesis of the supremal ABSRA, we present a synthesis algorithm, which ensures that a computed supervisor will be ABSRA-robust , i.e., either an ABSRA will be detectable or will not lead the system to an undesirable state.

Rong Su

To make a supervisor comprehensible to a layman has been a long-lasting goal in the supervisory control community. One strategy is to reduce the size of a supervisor to generate a control equivalent version, whose size is hopefully much smaller than the original one so that a user or control designer can easily check whether a designed controller fulfils its objectives and requirements. After the first journal paper on this topic appeared in 1986 by Vaz and Wonham, which relied on the concept of control covers, in 2004 Su and Wonham proposed to use control congruences to ensure computational viability. This work is later adopted in the supervisor localization theory, which aims for a control equivalent distributed implementation of a given centralized supervisor. But after so many publications, some fundamental questions, which should have been addressed in the first place, have not been answered yet, namely what information is critical to ensure control equivalence, what information is responsible for size reduction, and whether the partial observation really makes things different. In this paper we will address these fundamental questions by showing that there does exist a unified supervisor reduction theory, which is applicable to all feasible supervisors regardless of whether they are under full observation or partial observation. Our theory provides a partial order over all control equivalent feasible supervisors based on their enabling, disabling and marking information, which can be used to categorize the corresponding reduction rates. Based on this result we can see that, given two control equivalent feasible supervisors, the one under full observation can always result in a reduced supervisor no bigger than that induced by a supervisor under partial observation.