File compression is one of those ubiquitous technologies that you only notice when it goes wrong. Like when you want to watch the winning goal from last night’s final on your phone and it won’t download, or show wedding footage to your friends, not to mention that all-important Zoom meeting when the screen freezes at precisely the moment you have your boss’ full attention.
Smart as they were, Imperial College maths and computer science students Chri Besenbruch and Arsalan Zafar had never given it much thought either. That was until they had to transfer a gigantic packet of synthetic data and found they couldn’t. At the time Chri, and fellow student Arsalan, an actual rocket scientist (he has a degree in spacecraft engineering) were working on a study into the simultaneous mapping and localisation technology being developed for self-driving cars.
Not the kind of people to come across a problem without developing a burning desire to solve it, the pair began what turned into a 5-year journey deep into the world of data file compression. With zero background in the field, they found themselves wading into such esoteric reaches of advanced mathematics and statistics as the Shannon-Hartley theorem of information capacity and how it relates to Gaussian noise. What came out of that odyssey was a company, Deep Render, which, if Chri is to be believed, may just save the internet from complete collapse.
It is also a journey, says Chri, that they could never have completed without Innovate UK, who understood from the get-go that to get from where they were to a commercially viable product required deep research, and that was not going to happen overnight. The introduction came via their professors at Imperial College, London, several of whom sit on their advisory Board. They had worked previously with Innovate UK and knew how it could potentially help. They suggested a Smart Grant was what Chri and Arsalan needed to move their thinking forward. After receiving their first grant of £50,000 in May 2020, they were able to make the case for two further grants totalling £770,000 which helped Deep Render get to the point where they are attracting serious attention from the mobile phone industry.
Everything that is shared or sent via the internet or from our phones – it could be a film that we are watching on Netflix, or the sound of our voice when talking – first has to be turned into data. However, if it is going anywhere, that mass of data has then to be compressed into a size which can realistically be squeezed down the transmission “pipe”. This is done by means of complex algorithms that remove any data which is deemed non-essential or can be recreated on reception, so that what is actually sent down the pipe takes up less space and can squeeze through data bottlenecks and still be recognizable when it gets to the other side.
Chri was staggered by what they discovered: “What we found was that the compression algorithms in use are based on technology which hasn’t changed since the 1970s and that this first-generation technology hasn’t transitioned.” A lot has happened since then. In those days code was written entirely by hand and today’s AI techniques were still the stuff of science fiction.
Chri and Arsalan decided to go back to the drawing board and redesign a file compression technology from scratch. The idea was to use everything that is available now which wasn’t back then, particularly emerging technologies like machine learning and neural networks. “Our philosophy was basically if something is broken, we shouldn’t try and repair it. We should burn it.”
It was obvious to Chri and his partner Arsalan that this was a problem ideally suited to AI, a technology which is remarkably good at processing visual data. He explains why. “AI seems to evaluate images in much that same way as the human visual system.” In fact, the human eye is already the best data compressor known to humanity – with compression ratios at least 2,000 times better than anything developed to date. Just as our own visual system does not need to see every single little detail on someone’s face in order to recognize who we’re looking at, AI can learn pretty quickly how to fill in gaps in images with remarkable accuracy. AI is also a far more efficient way of generating algorithms.
Deep Render is now in active discussions with several big name US tech giants with a view to their adopting its compression technology as standard. This will mean Deep Render’s codec, to use the industry term for its combined coder and decoder, being pre-installed in every device they manufacture and sell worldwide. This would effectively hand Deep Render a global monopoly over one of the key processes behind every word, picture or film we receive or send digitally.
“This is a winner takes all game.” Unless everyone uses the same codec – devices won’t be able to talk to each other. There are three billion devices in circulation and manufacturers are currently paying $1.50 per device in royalties for the right to use the existing compression algorithms. So we are talking potentially eye-watering sums. “Once we have unlocked the commercial gate, it is game over.”
Deep Render’s case to be adopted as industry standard seems on the face of it utterly convincing. “Our Codec is 80% more efficient than the current standard,” meaning that a given piece of data can be compressed into a space five times smaller than at present for no loss of quality. With data usage doubling every two years that is a compelling argument. Adding in new fibre capacity is expensive, particularly outside densely populated urban areas, and with political pressure rising, governments are forking out massive subsidies to support roll-out in rural areas where the commercial case is marginal at best. New capacity is also only able to grow linearly, whereas data usage is growing exponentially year on year.
Chri is convinced that unless the industry adopts Deep Render’s technology or something very similar to it, it will soon be game over for the internet as we know it. Those who can afford to spend the massive sums needed to double and triple capacity year on year will invest, and charge those who are willing to pay accordingly. The vast majority of users will meanwhile be left stuck in the slow lane. It will kill off net neutrality, the cardinal principle of equal access that has governed the internet since inception.
Data usage is also a voracious consumer of energy, accounting for half of global CO2 emissions, and the amount is projected to grow by 17% by 2023. One can only imagine the impact of cutting that by 80% – which is what would happen if Deep Render’s codec was adopted worldwide.
“This is literally a must-have if the internet is to survive,” says Chri. Deep Render now has a team of 20 and has just been out in the market to raise the cash to support the next stage of growth. As a business, it has the potential to accrue massive revenue without needing much more in the way of resources to service that business. “An IP business scales even better than a software business,” Chri explains. What they need to do is land a few big whales, as a small number of potential customers would be sufficient to completely unlock the market. It is also a business with a uniquely valuable IP. So far it has acquired no less than 8 patents with a further 47 in the works.
Chri has nothing but praise for Innovate UK, and the help they provided at a crucial initial stage. The respect in which it is held within the scientific and venture capital fraternity gave them a lot of credibility when it counted. It tied their professors at Imperial much more tightly into the project. It also carried a lot of weight with other sources of funding. “Once we had a first believer, it was easier to find a second believer,” he says.
Born and raised in Germany, Chri studied in Germany and Switzerland before coming to Britain. He is therefore well placed to judge their relative merits as a base to create and build an IP-based business. Cultural and institutionally, the UK scores better, in his view, even than the US. Surprisingly, despite its reputation, he is convinced that in Silicon Valley they would have found it much harder to do what they have done than here in the UK.
For a start-up, attracting the right kind of talent would have been more difficult, not to mention more expensive given the Bay Area’s demanding talent to compensation ratio. Nor does the US offer the kind of public support for the kind of deep research-based tech that lay behind Deep Render’s success that is available in the UK. The UK has also become something of a “hotspot” for deep tech and AI, and investor here, generally speaking, have a better grasp of the kind of timelines involved.
Innovate UK may not have the firepower of some of the other tech funders in the US and Europe, but they more than made up for it in other ways. Not least their level of understanding of the science, and the ability to be rigorous in their evaluation and monitoring without stifling the creativity and entrepreneurialism needed to see the project through. “Where I come from, they simply don’t have the risk appetite. The UK is awesome.”
