Cyber Security - SWARM

so the merchant features or self-organizing systems um in other words swarming for cyber applications this is uh a talk that is going to be focusing not on specific techniques that we have available right now but more of a vision into the future I am Dr Sven Berkner I have my PhD in computer science artificial intelligence and I've been doing 15 years plus r d and small intelligence theory and applications cyber is one of those that have been moving into our company action go Sentinel part of the Mach 37 accelerator in the spring um markers are being a sponsor here of this conference we are looking to build uh cyber security products for mobile devices and

the internet of things and that is going to be a part of my presentation today um if you want to take an early peek at what we are doing uh our first uh Android app is out at the free public beta Hanukkah took it up if you like um part off axon AI which is looking at more than just cyber as applications for swarm intelligence for self-organizing systems so I'm going to be talking a bit about what swarm intelligence research is looking at and how that can be applied to mobile applications and email things we're going to sketch and approach that we are following right now so as I said self-organizing systems with emergent uh system level features

in other words swarming systems are all around us Nature has demonstrated in many many many indications on how it should take very simple entities and combine them adaptively of robustly scalably to create complex system level Behavior you have just got some photos there you've got the causal classical and systems of the bees swarming we've got the fish schooling to create uh the appearance of a much larger animal to protect themselves from predators we have swarming going on self-organization going on of course it at the bacterial level too but then at the other hand even humans in groups do IT Crowd Behavior Market Behavior those are all examples of swarming systems of self-organizing systems where uh at the

aggregate new Behaviors new patterns are emerging that are not encoded in the individual level Behavior so with all these examples around the question is why are we still programming our applications our Solutions in the traditional top-down centralized fashion why are we trying to spend and the enormous amount of time on knowledge engineering on um explicitly enumerating our potential conditions our system has to face uh when we can create complex Behavior by combining many lightweight individuals that are the autonomous autonomously acting in a shared environment that do not communicate directly with each other but through the manipulation of the environment the principle of the French biology because they called figma G that can deal with noisy inflammation

and a heterogeneous constantly changing population and out of all that to create a system at the system of the functionality emerges robustly adaptively uh at a large scale and has capabilities of learning um also building so when we look at engineered systems especially um in the internet of things because seeing uh in uh in architecture that is resembling all those natural systems that you're dealing with some uh the statistics that are being thrown around over there true or not the projections are by 2025 there will be more than 70 billion Internet connect Internet Internet connected devices those are simple devices uh that do their things whatever they're doing autonomously uh and um more than a trillion connected

sensors so if you're trying to build applications whether it's uh for uh securing your internet of things or whatever it is to provide those various functions that you are hoping to get out of DNA other things really you have a look at engineering um those devices into an aggregate into a coherent system by using uh self-organization and merging emerging as design principles I said I work on uh swarm intelligence really slow intelligence as a special case of artificial intelligence in the computer science World um takes its inspiration from those natural Asian systems and then we have two paths uh within the community you have from these observations and analyzes of natural naturally occurring self-organizations if people who are

focusing on building models of what happens in nature how can we explained that natural systems um create these functionalities how do ants form their paths or the termites create their air conditioning systems in their mouths and so on and so forth so models of the real world that are compared against the natural system observations but then the second half is more from the engineering perspective okay if you understand and have models on how nature creates self-organization organizing systems can be derived the basic principles that back design really takes as I was showing on an earlier slide things like keep your Agent Small do not have the right interaction and so on and so forth the laundry list of design principles

and apply them as an engineering methodology to create new systems so that's really where my work has been focused on looking at swimming robotics looking at planning and predicting complex Behavior looking at coordination of resources of various types and then finally looking at Cyber applications where you are tying together many different devices being the essential network will be at a local or a network of applications so how would we want to do that if we have a collecting a collection of physical devices in an infrastructure where the individual devices are limited in their capabilities in terms of bandwidth in terms of their computational power a nest thermostat does run Linux it is a little computer

but overall it's not going to be the power of the today's smartphones it's going to be more simple where we have devices coming and going all the times where we have the side churn that we have to deal with well if you're taking the perspective that uh you know let me call a device in our infrastructure is really the combination of a physical unit with an autonomous agent so a piece of software representing that unit to take any kind of control actions on that unit or take information from that physical units or kind of our interface plus provide that device with the ability to host other little software agents that could come and go through

just a standard massive capacity protocol if you have that kind of infrastructure we can then use those agents to create a self-organizing peer-to-peer overlay and create virtual topologies where uh rather than just thinking with the primary Communication channel channels that are available from one device to another route messages across the network of devices according to certain um objectives so I want to have devices of similar characteristics be tied close together in a peer-to-peer overlay or you can achieve something like that in this self-organizing fashion where devices basically search for other devices to connect to so as an example in that PHP are overlay we can then run other agents that are kind of spreading out information that are

taking local information and transforming it for the use of neighbors and neighbors of neighbors and so on in that fashion you can then create applications as emerging features of that process of these swarming agents just like you see in these natural agent systems colonies or other swamps that we find in Asia so one example of that um one example of that is our first application that we are working on right now so um say for instance we have um our mobile devices where we are taking local observations like okay what's the current CPU mode of the device or the memory consumption or ratio of inbound onboard traffic or other um in the indicators of device Behavior use

the self-organizing PHP overlay as one method to create a topology where devices with similar characteristics find themselves um connected to other devices that have the same same or similar characteristics so that you can Traverse that topology and as you're discovering devices that are suddenly stand out like a soil farming relative to all their neighbors find those identify those as suddenly changed as potentially uh abnormal behavior and stuff highlighting those so uh you can do all that with local choices with agents swarming in this topology that is created on the Fly by the network of devices and out of that you can get situational awareness you can know what type of devices are in your system right

now what are connected you can know what the characteristics of those devices are as well as detection of abnormal behavior potentially threatening behavior by looking at these abnormal or the deviations from from the Learned normal Network so that's the starting point for our development that we're doing at accent goes Sentinel and expanding from there adding more capabilities more features always 15's principles of Engineering self-organizing Systems in mind all your choices are local all your behaviors and rules are lightweight um so that you can scale up to a large number of devices in your network as well as deal with very weak devices and with that the short introduction is over so any questions comments out of curiosity

are there efficiencies gained in terms of resources in the Swarms versus a dedicated hierarchy uh well the advantage of using swarming rather than sort of self-organizing systems rather than explicitly coded hierarchies for one to deal with the dynamic situation we're creating a hierarchy if you try to do a top-down creates scaling challenges as well as just reactive challenges also the advantage of doing the swarming in the network and doing basically all your calculation all your operations in the network is that it's pushing in computation to the edge and thereby you're saving a bandwidth of not having to pump out all status information into the cloud or whatever and do an analytics there so that's really the

purpose of the swarming architecture push the calculation and operations into the cloud into the network to the action I I'm I do networking so I mean that's my primary jobs that's why I'm curious this has an assumption of a network right in place so it's more for it's less for self-organizing a network at this stage um you mean self-organizing the physical connections or wrecks right I mean the linkages between the nodes you can create that out of the swarming too I mean you can start with one set of linkages and then after swarming uh operating over that Set uh um basically identify okay which linkages should be adjusted uh and and recreated uh um so so that you get more

optimal traffic or that's where I was curious about what they would be over like I know I've got packing overhead wondering for a swarm based protocol yeah we don't have a working with us no it's a widely adopted model that's right I've been most interested seeing where with this type of research above and beyond what we can do with it robotics models or something do you have any suggested open source libraries or reading regarding swapping and definitely want to implement something about well um there are some packages out there most predominantly used or widely used as a repast which is a successor to a swarm which is an architecture that originally came out of Santa Fe I

believe we passed this uh maintained by University of Chicago Java based framework with lots and lots a lot of example models and if you just Google swarm intelligence uh as as keywords self-organization Americans you'll find plenty of reading there's the long-standing series of conferences called Sasso self-adaptive self-organizing systems that happens each year alternating between Europe and the United States that has good Publications entry level if if you were interested they typically have tutorial sessions in the first days so it's a good place to start

this is not lend itself to like biological calculations like manuals law and things like that as well talking is originally biologics um well it's not like it's not like uh genetic Computing or anything like that no no it's it's really um software engineering uh but uh bottom up rather than top down so it's it's a different methodology for software engineering even when you have an in robotics robotic platforms it's still the software that controls the robots and how to coordinate and all that um what's the kind of scale that you're looking at Alex you mentioned that you're there's a beta of a smartphone app is that is that going along the lines of where like my smartphone would

be one of the nodes in the Swarm and everyone else's that is Division I mean the current version of that is that the the Swarm is actually not physically running on your app in your app your app connects to the server where the Swarm lives so right now the agents are still sitting in a central server doing their swarming thing in that box uh but our next versions that we're working on is gonna more and more push um the boundaries where at the end of the day the the real division the vision is that there is no not going to be a central server anymore there is your device and my device and everybody else's device uh talking to each other

um supporting that swarming architecture so that uh we are creating the intelligence in the network amongst us there

what's good when I think of swarms I think of London's going on yes um traditional software um is brittle um to um even subtle changes if you hacked if if you hack into the software uh if you find a way in just like your talk before if you uh uh know the um hacking tricks to get into the software you can control the whole thing because it's it's it's a that the controllers at it's it's uh in in the software if you hack your way and you have control self-organizing system where even if you hack your way into the code it's not going to give you control of the system if you uh try to get hands

out of your kitchen they have invaded your kitchen they've built their paths killing the individual hand is not gonna do anything good you will have to basically wipe out most of the population and preferably also the the chemical markers the pheromones that they laid down so that they're guiding in new ads out from the colony so um from a from a vulnerability perspective self-organizing systems are or can be uh more secure uh because the individual the software itself is not going to tell you what the system is doing that's a self-organizing emergent feature of all the interactions uh taking control of an individual agent like okay I'm gonna install uh the uh app and then I'm gonna hack it and I'm

gonna send different signals that's just one among thousands and thousands of participants in the Swarm that's not going to be changing the Lemmings changing the overall Behavior but if you understand if you're basically um knowledgeable enough that you do understand what the system level behavior is you can start Faking It Out sending enough uh fake signals to affect many of the participants at some point you will be able to take control and send all the limits off the cliff but it takes a lot of effort to do that

any other questions thank you for your attention