Holistic Engineering Ltd Logo
Holistic Engineering helps you discover the real problem in its wider context
and collaboratively architects coherent ways forward.

Research & Development

Some challenges cannot be solved by improving today's answers because the questions themselves need to change. This is where our research begins. 

By applying the same thinking that underpins our consultancy, we explore and develop transformational technologies. Rather than simply optimising existing solutions, we ask whether the underlying assumptions are still valid.

One example of our current research and development work explores applications in two areas: 

  • Next generation / advanced compute for AI and information processing. 
  • Advanced audio processing with applications in hearing technology, compression and possibly more.

The Fundamental Problem

Current AI systems can be extremely computationally intensive because they repeatedly compare new inputs against vast amounts of stored information. This creates two major bottlenecks:

  • the computation required to perform these comparisons;
  • the movement of large amounts of stored data.

We concluded that a fundamental problem lies in how information is represented. This led to a different question: can the architecture be redesigned so that information is represented more efficiently and irrelevant information is ignored from the outset, reducing the amount of computation required ?

Solution

Rather than starting with a technology, we returned to the fundamental question and developed a new computational abstraction. This means the same underlying computational abstraction can be explored across different technologies and implementation domains. 

We have explored the underlying concept across the following implementation domains: 

If you want to understand some of the above in more depth, then have a look at these recorded lectures: 1 (Biologically Inspired Computing - The Future ?, 2016) & 2 (Quest for AI Systems, 2024). 

Visualisation of electronic signals on a board

Electronics

We designed electronics circuits that compare new information with previously learned information and tested them by identifying which instrument was playing a particular note.

Our current circuits are built largely from passive components, offering the potential for substantial power savings compared with existing approaches. Successfully performing these comparisons demonstrated that our computational abstraction can be implemented in electronics. 

During this work, we also discovered an unexpected benefit: our technology proved highly effective for audio decomposition - determining which frequencies are present in a signal at each point in time. 

Conventional approaches, such as the Fast Fourier Transform (FFT), typically require a trade-off between frequency resolution and time resolution. Our approach avoids this compromise by providing full spectral detail at every point in time, while also offering built-in redundancy. 

This has potential applications in: 

  • Hearing implants. Our initial results indicate that our audio decomposition method can retain around 10 times more signal information before it is presented to the brain compared with current approaches. If this translated into a clinical system, this could fundamentally improve the listening experience for hearing implant users. 
  • Audio compression. Our initial work demonstrates that we may be able to reconstruct audio from a limited information set, which would allow for compression and reproduction to full quality.

We are only at the beginning of this journey. If you are interested in a demonstration or would like to support our further work in this area, then get in touch. 

Optics

Optical Cross Correlation is not new, and offers significant advantages when identifying similarities between an input and stored information. Yet, despite these benefits, it does not appear to be used for image-based AI. 

One challenge is image storage, but we believe this could be overcome through further research and investment. 

Most existing optical approaches perform part of the computation optically before converting the information back into the digital domain. Our work takes a different direction. By applying the same computational abstraction described above, we have developed the foundations for a fully optical implementation. 

We would now like to take this work further and are actively seeking research collaborators and funding partners.

Visualisation of optical signals
Visualisation of quantum signals

Quantum

Based on what we have discovered through our work in electronics and optics, we believe quantum computing provides another promising implementation domain for our computational abstraction.

One potentially significant advantage of our approach is its inherent tolerance to noise. Because the underlying computational abstraction is designed to operate with noisy information, it may offer the potential to reduce some of the environmental constraints associated with certain quantum computing architectures, such as the need for extreme cooling to minimise thermal noise.

Exploring this remains part of our future research.

We are actively seeking collaborators and funding partners to help take this work forward.  

We believe tomorrow's breakthroughs will emerge from asking better questions, understanding problems more fundamentally and developing coherent, new ways of thinking. 

We are looking for collaborators, research partners and investors who share that mindset and want to help bring these technologies into the real world.

If this resonates with you, we'd be delighted to start a conversation.