Limited data availability in machine learning significantly impacts performance and generalization. Traditional augmentation methods enhance moderately sufficient datasets. GANs struggle with convergence when generating diverse samples. Diffusion models, while effective, have high computational costs. We introduce an Expansive Synthesis model generating large-scale, information-rich datasets from minimal samples. It uses expander graph mappings and feature interpolation to preserve data distribution and feature relationships. The model leverages neural networks' non-linear latent space, captured by a Koopman operator, to create a linear feature space for dataset expansion. An autoencoder with self-attention layers and optimal transport refines distributional consistency. We validate by comparing classifiers trained on generated data to those trained on original datasets. Results show comparable performance, demonstrating the model's potential to augment training data effectively. This work advances data generation, addressing scarcity in machine learning applications.1

In the era of big data, the sheer volume and complexity of datasets pose significant challenges in machine learning, particularly in image processing tasks. This paper introduces an innovative Autoencoder-based Dataset Condensation Model backed by Koopman operator theory that effectively packs large datasets into compact, information-rich representations. Inspired by the predictive coding mechanisms of the human brain, our model leverages a novel approach to encode and reconstruct data, maintaining essential features and label distributions. The condensation process utilizes an autoencoder neural network architecture, coupled with Optimal Transport theory and Wasserstein distance, to minimize the distributional discrepancies between the original and synthesized datasets. We present a two-stage implementation strategy: first, condensing the large dataset into a smaller synthesized subset; second, evaluating the synthesized data by training a classifier and comparing its performance with a classifier trained on an equivalent subset of the original data. Our experimental results demonstrate that the classifiers trained on condensed data exhibit comparable performance to those trained on the original datasets, thus affirming the efficacy of our condensation model. This work not only contributes to the reduction of computational resources but also paves the way for efficient data handling in constrained environments, marking a significant step forward in data-efficient machine learning.11Thanks to the generous support of ARO grant W911NF-23-2-0041

The power and flexibility of Optimal Transport (OT) have pervaded a wide spectrum of problems, including recent Machine Learning challenges such as unsupervised domain adaptation. Its essence of quantitatively relating two probability distributions by some optimal metric, has been creatively exploited and shown to hold promise for many real-world data challenges. In a related theme in the present work, we posit that domain adaptation robustness is rooted in the intrinsic (latent) representations of the respective data, which are inherently lying in a non-linear submanifold embedded in a higher dimensional Euclidean space. We account for the geometric properties by refining the l2Euclidean metric to better reflect the geodesic distance between two distinct representations. We integrate a metric correction term as well as a prior cluster structure in the source data of the OT-driven adaptation. We show that this is tantamount to an implicit Bayesian framework, which we demonstrate to be viable for a more robust and better-performing approach to domain adaptation. Substantiating experiments are also included for validation purposes.

We introduce in this paper a new statistical perspective, exploiting the Jaccard similarity metric, as a measure-based metric to effectively invoke non-linear features in the loss of self-supervised contrastive learning. Specifically, our proposed metric may be interpreted as a dependence measure between two adapted projections learned from the so-called latent representations. This is in contrast to the cosine similarity measure in the conventional contrastive learning model, which accounts for correlation information. To the best of our knowledge, this effectively non-linearly fused information embedded in the Jaccard similarity, is novel to self-supervision learning with promising results. The proposed approach is compared to two state-of-the-art self-supervised contrastive learning methods on three image datasets. We not only demonstrate its amenable applicability in current ML problems, but also its improved performance and training efficiency.

Expert networks are formed by a group of expert-professionals with different specialties to collaboratively resolve specific queries posted to the network. In such networks, when a query reaches an expert who does not have sufficient expertise, this query needs to be routed to other experts for further processing until it is completely solved; therefore, query answering efficiency is sensitive to the underlying query routing mechanism being used. Among all possible query routing mechanisms, decentralized search, operating purely on each expert’s local information without any knowledge of network global structure, represents the most basic and scalable routing mechanism, which is applicable to any network scenarios even in dynamic networks. However, there is still a lack of fundamental understanding of the efficiency of decentralized search in expert networks. In this regard, we investigate decentralized search by quantifying its performance under a variety of network settings. Our key findings reveal the existence of network conditions, under which decentralized search can achieve significantly short query routing paths (i.e., between O (log n ) and O (log 2 n ) hops, n : total number of experts in the network). Based on such theoretical foundation, we further study how the unique properties of decentralized search in expert networks are related to the anecdotal small-world phenomenon. In addition, we demonstrate that decentralized search is robust against estimation errors introduced by misinterpreting the required expertise levels. The developed performance bounds, confirmed by real datasets, are able to assist in predicting network performance and designing complex expert networks.

Malicious coresidency is a precursor to side-channel attacks that target information leakage. In this paper, we seek to understand the interactions between a defender (the cloud service provider) who tries to detect malicious coresidency by an attacker, who in turn attempts to co-reside its VM with a victim VM on the same physical machine by exploiting the VM allocation policy employed by the cloud service provider while at the same time, trying to evade detection. The problem is modeled as a two-player game. Specifically, the attacker chooses how long to keep its VM operational before terminating and relaunching it to increase its odds of success. On the other hand, the defender attempts to detect and penalize malicious VMs based on their activity in a given time window. The defender estimates a maliciousness measure for all active VMs which then modulates the likelihood of a specific VM being migrated to a different physical machine. We study the equilibrium strategies for both players for different ranges of environment parameters and show the non-existence of equilibrium with pure strategies. Subsequently, we characterize the equilibrium of the game with mixed strategies.

We propose a Context-aware Smallworld routing protocol for wireless ad-hoc networks which finds efficient routes to the destination nodes often through a very short chain of intermediate nodes, without using any routing table. This protocol exploits the “small-world” phenomenon of social networks where source-destination pairs typically get connected through few intermediate friends and the individuals can collectively discover such short paths. The key idea behind our routing protocol is that for forwarding an incoming packet to the destination, a node picks the next hop from among the one-hop neighbor from its one-hop neighborhood that is closest to the destination, and multiple randomly selected long-distance neighbor(s), i.e., from outside of its one-hop neighborhood, based on which of these neighbors is closest to the destination. The long-distance neighbors are selected by using their contextual information through a probabilistic mapping. We provide the theoretical foundations behind our protocol. This simple but effective algorithm can meet performance objectives for most scenarios by reducing loss and latency. Simulation results, using synthetic network topologies, have demonstrated that the routing performance of context-aware Smallworld is better than geographical (distance-based) Smallworld and other standard proactive routing protocols (e.g., OLSRv2) when metrics, such as packet losses, end-to-end delays, hop-counts, connectivity drops, and control traffic generated, are used as the key performance indicators. We also show that the proposed protocol is resilient to dynamic topology changes.

Incentive-driven advanced attacks have become a major concern to cyber-security. Traditional defense techniques that adopt a passive and static approach by assuming a fixed attack type are insufficient in the face of highly adaptive and stealthy attacks. In particular, a passive defense approach often creates information asymmetry where the attacker knows more about the defender. To this end, moving target defense (MTD) has emerged as a promising way to reverse this information asymmetry. The main idea of MTD is to (continuously) change certain aspects of the system under control to increase the attacker's uncertainty, which in turn increases attack cost/complexity and reduces the chance of a successful exploit in a given amount of time. In this paper, we go one step beyond and show that MTD can be further improved when combined with information disclosure. In particular, we consider that the defender adopts a MTD strategy to protect a critical resource across a network of nodes, and propose a Bayesian Stackelberg game model with the defender as the leader and the attacker as the follower. After fully characterizing the defender's optimal migration strategies, we show that the defender can design a signaling scheme to exploit the uncertainty created by MTD to further affect the attacker's behavior for its own advantage. We obtain conditions under which signaling is useful, and show that strategic information disclosure can be a promising way to further reverse the information asymmetry and achieve more efficient active defense.

We study the problem of network topology design within a set of policy-compliant topologies as a game between a designer and an adversary. At any time instant, the designer aims to operate the network in an optimal topology within the set of policy compliant topologies with respect to a desired network property. Simultaneously, the adversary counters the designer trying to force operation in a suboptimal topology. Specifically, if the designer and the attacker choose the same link in the current topology to defend/grow and attack, respectively, then the latter is thwarted. However, if the defender does not correctly guess where the attacker is going to attack, and, hence, acts elsewhere, the topology reverts to the best policy-compliant configuration after a successful attack. We show the existence of various mixed strategy equilibria in this game and systematically study its structural properties. We study the effect of parameters, such as probability of a successful attack, and characterize the steady state behavior of the underlying Markov chain. While the intuitive adversarial strategy here is to attack the most important links, the Nash equilibrium strategy is for the designer to defend the most crucial links and for the adversary to focus attack on the lesser crucial links. We validate these properties through two use cases with example sets of network topologies. Next, we consider a multi-stage framework where the designer is not only interested in the instantaneous network property costs but a discounted sum of costs over many time instances. We establish structural properties of the equilibrium strategies in the multi-stage setting, and also demonstrate that applying algorithms based on the Q-Learning and Rollout methods can result in significant benefits for the designer compared with strategies resulting from a one-shot based game.

Expert networks are formed by a group of expert-profes\-sionals with different specialties to collaboratively resolve specific queries. In such networks, when a query reaches an expert who does not have sufficient expertise, this query needs to be routed to other experts for further processing until it is completely solved; therefore, query answering efficiency is sensitive to the underlying query routing mechanism being used. Among all possible query routing mechanisms, decentralized search, operating purely on each expert's local information without any knowledge of network global structure, represents the most basic and scalable routing mechanism. However, there is still a lack of fundamental understanding of the efficiency of decentralized search in expert networks. In this regard, we investigate decentralized search by quantifying its performance under a variety of network settings. Our key findings reveal the existence of network conditions, under which decentralized search can achieve significantly short query routing paths (i.e., between O(log n) and O(log2 n) hops, n: total number of experts in the network). Based on such theoretical foundation, we then study how the unique properties of decentralized search in expert networks is related to the anecdotal small-world phenomenon. To the best of our knowledge, this is the first work studying fundamental behaviors of decentralized search in expert networks. The developed performance bounds, confirmed by real datasets, can assist in predicting network performance and designing complex expert networks.

Expert networks are formed by a group of expert-professionals with different specialties to collaboratively resolve specific queries posted to the network. In expert networks, decentralized search, operating purely on each expert's local information without any knowledge of network global structure, represents the most basic and scalable routing mechanism. However, there is still a lack of fundamental understanding of the efficiency of decentralized search. In this regard, we investigate decentralized search by quantifying its performance under a variety of network settings. Our key findings reveal that under certain network conditions, decentralized search can achieve significantly small query routing steps (i.e., between O(log n) and O(log2n), n: total number of experts in the network). To the best of our knowledge, this is the first work studying fundamental behaviors of decentralized search in expert networks.

We investigate the problem of query answering in expert networks, which are composed of inter-connected experts with various specialties. Upon receiving a query, the expert network is tasked to route this query to experts with sufficient expertise in a timely and reliable manner. However, the efficiency of query answering depends on the underlying query routing protocol being used. Among all possible query routing protocols, decentralized search, operating purely on each expert's local information without any network global knowledge, represents the most basic and scalable routing protocol. However, there is still a lack of fundamental understanding on the efficiency of decentralized search in different expert networks. In this regard, we establish a generic model that can abstract diversified social and structural attributes in various expert networks into a common framework, thus applicable to a wide range of network scenarios. On top of such generic network model, we then study decentralized search by quantifying its performance under a variety of network parameters. Our key findings reveal the existence of network conditions, under which decentralized search can achieve significantly short query routing paths (i.e., between O(log n) and O(log2n) hops, n: total number of experts in the network). To the best of our knowledge, this is the first work studying fundamental behaviors of decentralized search without relying on strict underlying network structures in expert networks. Experiments in both synthetic and real expert networks confirm the efficacy of the developed performance bounds in understanding and reasoning the network performance.

We study the problem of network topology design within a sequence of policy-compliant topologies as a game between a designer and an adversary. At any time instant, the designer aims to operate the network in an optimal topology within this policy compliant sequence with respect to a desired network property. Simultaneously, the adversary counters the designer trying to force operation in a suboptimal topology. We show the existence of various mixed strategy equilibria in this game and systematically study its structural properties. We study the effect of parameters, and characterize the steady state behavior of the underlying Markov chain. While the intuitive adversarial strategy here is to attack links appearing early in the topology sequence, the Nash Equilibrium strategy is for the designer to defend the earlier links and for the adversary to attack the later links. We validate these properties through two use cases with example sets of network topologies.

The landscape of cyber security has been reformed dramatically by the recently emerging Advanced Persistent Threat (APT). It is uniquely featured by the stealthy, continuous, sophisticated and well-funded attack process for long-term malicious gain, which render the current defense mechanisms inapplicable. A novel design of defense strategy, continuously combating APT in a long time-span with imperfect/incomplete information on attacker's actions, is urgently needed. The challenge is even more escalated when APT is coupled with the insider threat (a major threat in cyber-security), where insiders could trade valuable information to APT attacker for monetary gains. The interplay among the defender, APT attacker and insiders should be judiciously studied to shed insights on a more secure defense system. In this paper, we consider the joint threats from APT attacker and the insiders, and characterize the fore-mentioned interplay as a two-layer game model, i.e., a defense/attack game between defender and APT attacker and an information-trading game among insiders. Through rigorous analysis, we identify the best response strategies for each player and prove the existence of Nash Equilibrium for both games. Extensive numerical study further verifies our analytic results and examines the impact of different system configurations on the achievable security level.