Hello, I'm Louis Passfield, I'm Professor of Sports Science here at the University of Kent and I'm also a member of the Endurance Research Group. Over the last 25 years I've been working in sports science and I came into sports science with a particular plan. My aim was to win the Tour de France and I thought by studying the sports science and applying it to my own training I'd be able to be successful in this way. Instead I learnt two important lessons from studying sport science, the first was that science doesn't yet have all the answers and the second was that unfortunately I lacked the talent to win the Tour de France. So my ambition had been to be able to stand on top of the podium but unfortunately I realized that was never going to be the case, not as a cyclist anyway. However, I've been very privileged over the course of my time as a scientist to work in a number of elite environments and in particular I was able I was part of the team that prepared for the Barcelona Olympics, the Atlanta Olympics and also the Beijing Olympics too. And more recently instead of working with the athletes directly I've been mentoring members that other scientists or members of those teams working for the London Olympics and now for Rio, working with the English Institute of Sport. Cycling is a wonderful sport for scientists because we can take measurements and calculate things. These particular devices are a power meter, which measures exactly how hard the cyclist is working and we can attach this directly to the cyclist's bicycle so we can see how hard she he or she is working during their training or their races. We first used this device to conduct a study looking at the hour record and this is a record conducted on the track and what we do is we see how far someone can cycle in one hour and the harder you work, the further you'll go in that hour. If you look at this graph here you'll see records set from the late 1870s right through till the 2000s and in particular notice how the record increases rapidly in those last few years. Now we were intrigued by this rapid increase in performance and what the cause for that might be and so what we tried to do was use data from the power metres to model how the record was broken and what the reasons behind this were. One of the complicating factors for this was that the records were broken using different cycling positions. So here you'll see the traditional racing position that was used for many of the records up until the beginning of that increase and then the position evolved into this one where more aerodynamic position and aerodynamic helmets and clothing was adopted. And then along came the Scotsman Graeme Obree who innovated even further, first by adopting a position which sawed his arms off and he bent forward and cycled like this. That position was subsequently outlawed by the International Cycling Federation and so he innovated again and developed a Superman position with his arms out in front of him instead. Subsequently, we were to calculate that both of these positions were much more aerodynamic than anything that had preceded it. So when we came to look at the hour record and do the calculations what we could do what we could see was how hard it each of the riders had worked in order to break the hour record. The records coloured in red here show where the hour record was broken with a lower power output, in other words the rider was not working quite as hard and the record was that was developed by innovation rather than by changes in training or fitness. So you can see every other record in the run through that period was developed by increases in cycling position, aerodynamics and equipment, rather than through changes in fitness. Next it went on to look at pacing or the way in which you ride during a race and the strategy you adopt. Here we see a course where we have a hill, or two hills I should say, and also periods of headwind or tailwind, so the arrows mark the direction in which the wind is blowing. The challenge for the cyclists then is to work out how to distribute his effort during that race, if the course were flat and the wind changed the wind didn't change direction then a uniform effort all the way through would be this the most successful strategy. But, when you have hills, uphill and down, and head winds and tail winds, what then do you do? Do you work harder into some sections, easier into others and how much should you vary your effort. Again, we were able to use mathematics in this situation to model what happens to the rider during this race and how that effort should be distributed. In this case, we can use this relatively straightforward equation, a linear equation of motion for the cyclists, to calculate the impact of different strategies and by comparing those different strategies we were able to show that over the course of a 40 kilometer race it was possible for a cyclist to save 30 seconds by varying the effort, working harder in the slower sections of the race, so by going uphill and working harder or into the headwind and working harder, you actually reduce the amount of time you lose there and then you can make it up by going more easily into the downhill sections and into tailwind sections and overall then the race time is improved. Subsequently, I worked with Patrick Cangley at the University of Brighton to develop a more complicated model. So the last one was simply the equation you saw in the slide, this one, what we did was we took the whole bicycle and the rider and put them into the computer completely. With this model we can change any aspect of the of the bicycle or the rider, we can change the wheels, the tyres, we can make the rider large or small, change the weight, and we can run the model over any course that can be that can be captured from, for example, Google Earth. Using this model, you can pedal it so that if you don't ride correctly it will even fall over and what we'd ended with, was then use an actual race to feedback that model to the riders in real time. Here on this graph you can see the profile in the dotted line here of the course. So we've got a big hill in the middle of the course and a descent towards the end and then we also showed the power profile or how hard the rider should work during the race. What we then did was tell the rider, during the ride itself, how hard to pedal. So that you can see that as they as the hill goes up, as a rider ascends, the power output is increased and then, as the rider hits the descent, power output decreases during the easier section of the race. In this actual scenario what happened was the rider was able to save 12 seconds over his normal performance by following the model, rather than trying to adopt a constant pace throughout. So now we know not only that this model works in theory but it works in practice as well. In future, what we imagine is that something like this can happen. These glasses are commercially available now and they will give you feedback in real time on how hard you're working and so it would be possible for us to provide the strategy for the rider on their glasses in real-time as they race. Now, clever as that model may be and the technology that's gone into it, one of the key things is that we still don't know how to train the model, we still can't that tell it what the most effective exercise it should be doing to enhance its performance and that's the bit that I'm really excited or really passionate about. So in fact we can calculate or measure the training input and we can look at the consequences of that training in terms of the output or the performance, but what we don't know is how those two are linked. This bit in the middle, the training, is still very much a black box. So anybody who's working with athletes or even people who are exercising for maximum health benefits are really relying on the scientists and their intuition or experience in order to prescribe training, it can't be done from a scientific basis yet because we simply don't have the information. Now one of the things that I'm particularly interested in is whether we can use the data that we're gathering from training to learn more about the process of training itself. So devices like this are able to capture huge amounts of training data and perhaps we can then interrogate that data to learn more about the training process. So this is the challenge that I'm working on currently. The difficulty here is that training is not easy to model. This is an example of what you get if you detach a power metre to a bicycle of a reasonably elite rider, this is a four hour training session and you can see how the effort varies rapidly throughout the ride and you can see there are periods of high-intensity work here and here, big blocks where the power outputs are quite high, but overall the pattern is what we'd call stochastic or hugely variable, virtually impossible to use for mathematical modeling. The other thing that we don't know is which bits of this training are the most important. So, for example, some scientists have suggested that just the most high-intensity training, conducted for only short periods of time is enough, so just what's in the yellow band here. Other scientists and quite a number of coaches suggest that polarized training, where you combine high intensities, so this yellow band at the top, with low intensities is the best way of training effectively. But these bands and where those lines are drawn are entirely arbitrary, they're not scientifically derived. So maybe they should be here, or perhaps a little higher, or maybe a lot lower, we really don't know the answer to these questions at all from a scientific perspective at the moment. So perhaps what we can do is turn this around and study the training of athletes in order to learn more about that process. In this particular study, what we did was look at the training of well-trained endurance runners. We took a large dataset gathered by GPS and simply put watches on the wrists of runners in order to get that data. And what we were trying to do is to identify which periods of the training were effective or ineffective and essentially to try and learn as much from that data as possible. We had lots of data to learn from, so the first thing we had to do was to take an individual training session, such as this one, and try and work out a way of modelling one session. Once we've got one session then we could think about what we would do with a week's worth of training or a month's worth of training or even a year's worth of training and capture all of that and try and interrogate that too. And then of course we had more than one runner, we had 13 runners in this study, so we had to do it for all 13 runners too. The major challenge we had was trying to make sense of those individual sessions and we were very pleased that we were able to come up with a solution for that, which we've called the training distribution profile. These two squiggly lines at the top here show two different training sessions for a runner. These are very difficult to model mathematically because of the huge variability, the stochastic nature of them that I talked about earlier. Beneath them though is exactly the same data, all of the information still present, but now transformed into a training distribution profile. This is something that's very easy to describe mathematically and now we can chop that up mathematically and ask the question, which bits of that training are the most effective. And by using these training distribution profiles what we were able to do is to take that group of runners analyze the whole of their training for a year and here's an example of one runner's training shown for the whole of the study on one plot, which now it looks a lot simpler, a lot cleaner than the previous graphs we were looking at and by analysing that data we were able to show that using the training profiles we could both visualise the data but more importantly we could calculate from their data which speeds mattered most. And for the cohort as a whole what we found was that their training between five point three and five point seven meters per second was the training that was associated with increases in performance, so, if you like, this was the magic bullet for those runners. We also found that their laboratory tests that we conducted told us nothing useful in terms of predicting which training intensity they should do. So by analysing their training rather than their laboratory measures we were able to learn more about what enhanced their performance. And finally, we could use the model we developed to actually predict how much their performance would improve. So we could link how much training they did to how much their performance would benefit too. So, in the future, we can imagine a situation where athletes compete not by seeing who crosses a line first but rather riding or running next to each other side by side and then at the conclusion of the race turning tapping each other on the shoulder to download the data and then comparing their performances through the data itself. Thank you