PAIMEMS
Polymorphic Artificial Intelligence Management and Experiential Memory Systems
📋 Read Technical Abstract | Part of the TRIPLE Patent Family (US Patent 11,956,352 B2)
AI That Keeps What It Learns
The pointers make the network polymorphic. Because its neurons are implemented by pointers that link a share of processing capacity to a share of working memory, the network can be reshaped in software: mapped through randomized configurations, grown, pruned, swapped for a stored version suited to the task, or moved clear of damaged hardware.
The Market Opportunity
Of that total, $1.48 trillion is forecast for AI infrastructure: the chips, servers, networks, cloud capacity and devices that AI runs on. At that scale, losing what a system has learned, and paying for capacity a network does not use, are expensive habits. PAIMEMS targets both.
The Problem: Lost Experience and Idle Hardware
🔌 Experience Lost at Restart
When a system shuts down or reboots, what its AI learned during that session is lost unless something records it and hands it back.
⬆️ Experience Lost at Upgrade
When a model is replaced or upgraded, what its predecessor learned in service stays behind unless something carries it across.
🧱 Fixed Networks Hold Idle Capacity
A network whose structure is fixed keeps all the neurons it was built with, and the memory they occupy, whether or not the task still uses them.
⏸️ Change and Damage Mean Downtime
Restructuring a network, or working around failed hardware, usually means stopping the network or keeping spare hardware on standby.
📡 Models Copied from Their Emissions
University researchers used a probe on a chip running an AI model to recover the model's design, layer by layer, with 99.91% accuracy from its electromagnetic signals, without hacking into the device (North Carolina State University, 12 December 2024). The attack compared those signals with an identical device's, so a network with a fixed layout gives such an attack a steady target.
How the Learning Loop Works
as it works
working data
- A network works. During a session, a neural network built from pointers runs on the system's processors and working memory.
- It meets new data. Readings from sensors, other inputs, and feedback on the results of its own actions arrive while it works, failures as well as successes.
- Learning is recorded as it goes. What the network learns from that data is written to storage automatically, during the session.
- A later network runs. In a later session a network starts work on the same system: the same one after a restart, a different one, or an upgrade.
- Experience becomes working data. The recorded learning is supplied to that network as data it operates on, so it can perform the task differently.
Polymorphic AI: Networks Built from Pointers
🌱 Grows and Prunes as It Works
Neurons and connections are created by adding pointers and removed by deleting them. The processing capacity and memory of a removed neuron are released for other work.
🔄 Changes Without Stopping
The network's structure can be modified while it continues to operate. The structure it grows can be recorded, then restored for a later session.
🎯 The Right Network for the Moment
A library holds stored networks suited to different tasks or operating conditions. A call, a sensor reading, feedback or another input determines which one to load, and the resources of the network it replaces are released. When a vehicle's sensors detect rain, for example, it can switch to a network suited to wet roads.
🛠️ Moves Clear of Damage
If a processor or memory module is damaged, it can be isolated and the affected pointers reassigned to working hardware, so the network can carry on while enough resources remain.
🎲 A Fresh Layout on Restart
A functionally identical network can be mapped to a fresh, randomized arrangement of processors and memory when it restarts. From outside, the same network is very unlikely to look the same twice, so signals gathered on one occasion may be useless against the next.
⚛️ Quantum-Coupled Networks
The method extends to quantum computing: quantum bits can be created and coupled to an operating neural network.
Where PAIMEMS Applies
PAIMEMS suits systems where AI should keep what it learns, make full use of its hardware, or keep working under pressure.
🤖 AI Assistants and Agents
Assistants and agents built on PAIMEMS pick up where they left off, with experience carried across sessions, restarts and upgrades.
🚗 Autonomous Vehicles and Robots
Networks chosen for the conditions their sensors detect, such as rain, with experience kept from one journey to the next.
🛡️ Defence and Critical Systems
Systems that can keep operating around damaged hardware, and present a different electronic signature when they restart.
📱 Edge and Embedded Devices
Devices that load only the network the task needs and release the rest, making fuller use of limited processors and memory.
🏢 Data Centres
Networks that prune what they do not use and release capacity to other work, aimed at higher utilization and lower running costs.
🔐 Model Protection
Randomized mapping designed to resist copying a model through side channels such as its electromagnetic signals.
Questions
What does the patent application cover?
The learning loop: while a neural network built from pointers works, what it learns from new data is recorded, and in a later session that learning is supplied as working data to a network running on the same system. The Polymorphic AI capabilities above build on it. US Application No. 19/801,919 is pending.
Which networks can use the recorded learning?
A network running on the same system in a later session: the same network after a shutdown or restart, a different network, or an upgraded one. The learning is supplied as data for that network to work from.
How does PAIMEMS relate to TRIPLE?
PAIMEMS belongs to the TRIPLE patent family. TRIPLE (US Patent 11,956,352 B2) secures communications with transient languages built from pointers; PAIMEMS applies pointer-built structures to neural networks and the experience they gain.
How does an engagement begin?
PAIMEMS is a patent-pending design, available for licensing, joint development and investment. Tell us about your platform using the form below.
For Investors and Licensees
📈 For Investors
A patent-pending position on AI that keeps and reuses its experience: neural networks built from pointers that record what they learn and hand it to later networks. Worldwide AI spending is forecast to reach $2.7 trillion in 2026 (Gartner, 16 September 2026). PAIMEMS sits in SDI's TRIPLE patent family, which includes granted US Patent 11,956,352 B2.
🔑 For Licensees
License PAIMEMS to build AI products whose experience survives restarts and upgrades, whose networks reshape themselves to the task and the hardware, and whose electronic signature can change when they restart.
🤝 For Development Partners
Bring Polymorphic AI to your processors, devices or platforms with SDI as your development partner.