 I don't have any slides or anything. I just thought rather than have a lightning talk session with only one talk, I would come up and say something about my research. So my name is Robert Miles, I'm a PhD student at the University of Nottingham, and I'm working on artificial immune systems. So if anybody interested in machine learning, cool interesting algorithms, I think my research is really cool. So I want to talk about it. So yeah, a lot of people don't even know that this field of artificial immune systems exists. The idea is if you're familiar with artificial neural networks, these are algorithms which are designed to copy the way that neural networks work in the brain. You study the biology of the neurons and you figure out some kind of simplified mathematical model of how neurons work, and then you can generate, you make a computer simulation of those neurons, and you can then train them to solve tasks, and these are quite commonly used tools now in machine learning. And the idea of artificial immune systems is to do the same kind of thing where you steal from biology, you imitate biology, whatever, but copying the human immune system, which does not get nearly enough credit, the human immune system is amazing. Biologically, it's a fantastic system, and in fact it qualifies by most definitions as a cognitive system. If you think about what the immune system is capable of, it's able to recognise that it's under threat, it's able to recognise threats that it's never seen before, newly-evolved bacteria, it will recognise them, it will identify a way of responding, and it then develops that response in real time, which is an unbelievable achievement that people don't think about. Bacterial generations are extremely short, they can out-evolve us immediately, so how are we not just being killed by viruses and bacteria constantly? We have systems that can learn and adapt in real time. Those systems also have memory, which is another key component of a cognitive system. You get chicken pox once as a child, your system learns how to beat it, remembers how to beat it, you can't get chicken pox again. It's also very robust. In machine learning there are two types of mistakes you can make, the false positive and the false negative. In the human immune system, a false positive is something like an allergy or an autoimmune disease, where the immune system registers something as a threat, which isn't really a threat. That can be annoying, like hay fever, or it can completely kill you. False negatives, similar situation, the immune system failing to register on something which is a problem. If that happens, you are dead. Either of those situations is very serious. The immune system has developed to be extremely effective at robustly classifying things into dangerous and not dangerous. Maybe we'd like to be able to steal this capability. Maybe we'd like to be able to imitate it. The most obvious application is computer antivirus. If we could have an antivirus that works the way the human immune system works, the problem is essentially solved. It's not commonly known the way that current antivirus works. It's basically just a list. Everything that has been identified as a threat is added to a list. You download the list regularly and check for the things on that list. It's generally not much cleverer than that. I think we can do better. That's the idea of artificial immune systems. Study the biology. Steal the mechanisms of action. Create computer algorithms that behave in that same robust way. The algorithm that I'm working with is called the dendritic cell algorithm which works by copying a specific type of cell in the body, the dendritic cells, which are sort of like little detective cells effectively. They actually move around the body and they collect a lot of information about what's going on, what antigen is present basically in the tissue, which is anything that's sort of biological molecules of various kinds. Bacteria will produce, in the process of their metabolism, they produce certain chemicals which it looks for. It also looks for signals produced by the cells themselves. So healthy cells will produce certain chemicals that indicate that the cell is happy and healthy and operating in the way that it should be. If a cell believes itself is under attack, it's under stress in some way, it's under threat of destruction, it releases certain other signals, chemical signals, which are sort of distress signals. By moving around and collecting these two things, what's going on and is the tissue okay, it's then able to correlate those two together and answer the question of what is causing a problem, right? So if these cells are moving around and they come across some particular molecular signature that they haven't seen before, they don't necessarily trigger the alarm if there's no damage. If all of the human cells are perfectly happy, then the immune system isn't triggered. This is why... See, this is what happens when you try and do a talk with absolutely zero planning. I should have gone back and explained the danger model versus the self-non-self model. Basically it's a relatively recent change in immunology. People used to think that the immune system detected things in your body that didn't originate there. That's the most simple idea, right? They go around, they check, is this part of my body or is it not? It turns out that that's not how it works because there's all kinds of things living in us which are actually no harm to us. Things like our gut flora or various other semi-symbiotic organisms and microorganisms that aren't causing us any problem and therefore don't justify any immune response. It turns out the way that it works is by detecting damage which is what I was saying before. I went on a digression and now I'm lost. Help me out here. What was I talking about? Before I took that diversion, what was I talking about? Dendritic cells. They go around the tissue and they collect this information and then if they perceive enough danger they report back to a lymph node where they dump everything that they've seen. They say, right. I was just... They report in saying I was just in some tissue which I'm telling you now was under attack. I don't know what was carrying out the attack but here's everything I've got. Here's everything I collected while I was going around the tissue and the lymph node collects that and then if it gets enough dendritic cells coming in and saying I was in tissue that was under attack here's what I've found and they've got something in common. There's a particular kind of smoking gun that is found in all of the cells that came back from damaged tissue and not found in the cells that came back from healthy tissue is then able to identify it and pretty much just hand that off to the T cells which are the vast... I mean they're tiny obviously but relative to other cells they're vast the killer T cells, right which are the things that hunt down and envelope and destroy invading bacteria. Obviously this is massively oversimplified but this is the basics of how it works. So yeah, the idea then is from that model we've designed and built an algorithm that works in a simplified way that you can feed information from a computer system in a computer security context you sort of build a big analogy it's a big metaphor where the computer is a body and the processes are the processes actually the computer processes are the biological processes it's kind of quite a tight analogy in that most of the names are the same, right you're still trying to stop viruses and long story short it works fairly well it's still very early stage it can't be released how long have I been talking for by the way? I don't have a sense of that Oh yeah there's nobody I'm not making anybody else late yes anyway this is cool research is what I'm saying it's kind of interesting and a lot of people don't know that it's even a field so I'm letting you all know artificial immune systems they're a lot of fun yes, questions it does oh right, yes sorry so the question and correct me if I'm misinterpreting this is if the body is able to learn for example how to defeat chicken pox and then we'll then be able to defeat it again immediately if it's re-exposed why doesn't this go into our genome so that we actually evolve to permanently be immune to chicken pox and basically the answer to that is I'm not actually an immunologist I know enough immunology to do the computer science but my understanding is that there's a vast vast difference in the speed that we can evolve compared to the speed that bacteria can evolve and viruses can evolve because we are very large animals that live a very long time we can't reproduce at all for decades after we're born so and also because we're sexually reproducing animals there's a limit to how much mutation we can have in any one generation because you still need to be compatible so it means that we evolve very slowly whereas bacteria if there's a single bacterium gets a single mutation which is helpful it can divide 10,000 times and a single generation I don't know what it is but it can be in optimal conditions it can be the course of hours for it to split and then split again and split again so because bacteria evolve so much quicker than we do any defences that we manage to evolve they can immediately counter in relative time scales because they just out-evolve us and the fact that the human immune system is so clever is the only way that we actually manage to the only way that we actually manage to survive at all under the constant onslaught that if you leave something out you get some meat leave it out on a plate, whatever now long until that meat is completely destroyed and crawling with parasites and bacteria not very long, right? and in principle the only thing standing between us and that is our immune system every single day you're fighting off all of that attack that would happen to that piece of meat skin is handy but if you leave the skin on the... also the skin is an immune organ like it is part of the immune system but even if you leave the skin on a piece of meat it's still going to go pretty quickly so the answer is that they just... they just devolve quicker than we do yep right, they physically go there it's kind of cool there's a protein on the surface which causes them to causes them to bind to other cells and the extent to which they express this protein increases by the amount of danger that they've seen and then once it reaches a certain threshold they will stick to a nearby cell and they sort of roll towards the nearest lymph node because on each cell there's kind of a chemical gradient of... I can't remember what it is but there's a chemical gradient towards each lymph node, the nearest lymph node and so once they stick they gradually sort of roll towards it and then when they get there I believe they actually... they're actually destroyed they just sort of open up and spill everything out into the lymph node along with the... yeah it's kind of sacrificial it's pretty metal actually but you know they're not conscious so who cares yeah oh sorry hi go next right yes not easily I'll try so there were kind of a couple of parts to that question and I've now forgotten them can you like bullet points okay I'll do the second one because I know the answer to it so the question is how easy is it to find a damage analogue that will work in some other context than the body right so what you're saying in a machine learning context and yeah it is kind of a problem so what's what's commonly done in the computer security example that I gave before there were earlier it actually combined with existing antivirus systems that were looking at experimental antivirus systems that were looking at things like network traffic patterns that for example like a worm will produce a particular pattern of network traffic when it's doing a ping scan for example to try and find other hosts on the network to infect and so if you have something that can detect that there are a lot of these sensors which you have to just put a threshold on you say right how likely is it right now that there's a ping scan going on from this host and the existing systems there's just a threshold if it exceeds a certain level then somebody gets an email saying we think somebody is attacking our network and then the human being can respond whereas with this, with the dendritic cell algorithm you can take that output from the network monitoring tool which is continuous or one of the danger signals you sort of put them into a matrix and just multiply them all by some constants according to how important you think they are you weight them and it will then be able to tell you not just is a ping scan going on but which process is responsible and then you have the capacity for an automated response because you don't need to send an email to someone you can just send a kill signal to the process and kill it and you strongly suspect that it's involved in some kind of attacking behaviour but yeah in general, figuring out the damage analogue is one of the harder parts of building the analogy yeah exactly and you can find those programmatically if you build a big data set in which you deliberately run some attacks and then compare and do PCA on it and find the ones that are most important I don't know there haven't actually been very many applications of this yet other than this one but yeah that's a big open area that people are looking at yeah right the question is we're taking from biology and going to computer science could we potentially go the other way man I hope so really not my department but yeah and this is why actually the field is called artificial immune systems there are journals for artificial immune systems and I hate the phrase artificial immune systems I think it's rubbish because it's deeply confusing to people like I have more than once now received messages from people who have read online that I'm working on artificial immune systems and they or some family member has some horrible auto immune disease and they want to know if I can help and it's just the most it's so heartbreaking to be like no I'm just making antiviruses and stuff nothing not only can I not help you but I don't believe there's anyone who can do better than the current immunologist so talk to your doctor sorry so I prefer to call it immune inspired computing because that's really what it is I can understand where people came from with that because an artificial neural network is really doing the work of a neural network but an artificial immune system is just simulating one because it's not a purely information based process you actually have to physically kill the bacterium or whatever but yeah in the future molecular nanotechnology will totally do it 2045 that's hard day January 1st 2045 we'll have it yes that's a really good question so why can we learn how to defeat chicken pox and we can't learn how to defeat other things with our immune systems right is that a fair assessment of what you said yeah I don't know again that would be great if we could figure out if we could figure out what it is about particular about that range of attacks that we can detect and remember and respond to appropriately that would be like a huge advance in medicine in general but no I don't know and I'm not certain that anyone does but don't quote me on that yeah so I can't really hear you sorry yes so there's almost nothing that can't be used for evil I mean almost anyone's research so the question is basically can this be used for surveillance and yeah oppressive whatever I mean so I'm a I'm a card carrying member of the open rights group I care about this stuff I've got to say my research I'm not concerned about it there's a whole load of machine learning stuff that has really big and serious applications in that field and there was a talk about that wasn't there about privacy I remember reading about one like big big data and its consequences for privacy or something there are people in my department working on some stuff that does creep me out but not my work I don't see that I don't see the application to be honest we're just going to run up a bit sir can we have your name please Robert Miles can you put your hands together for Robert Miles he's come out of the blue and he's been fantastic