TIKOS®

"What makes you different and why should I care?" Tech obsessed founders will spend way too much time swimming in their secret sauce and forget others don't care about the details. I was caught out this week which prompted the following addition to our presentation. What Makes You Different? TIKOS® are the only company who can access and analyse internal decisions of deep-learning models, including open-weights LLMs. Why Should I Care? Without this granular understanding and control you risk millions in potential fines, litigation, failed projects and competitive threats. (but if you are tech curious, see it in action: https://lnkd.in/eg7YN5-t) #KeepItSimple #RiskManagement #FinServ #TrustworthyAI

Excited to share the latest work from TIKOS®, presenting a groundbreaking method for identifying and eliminating bias in AI models. For too long, technical, product and risk teams have had to rely on correlation tools, leaving root causes of bias undetected. This article introduces a new approach that provides hard evidence, allowing us to pinpoint and surgically fix bias in Deep Neural Networks at the neuronal level. This is a must-read for anyone that deals with AI bias, fairness, model risk and is especially relevant for those in financial services where the requirements of the FCA, PRA and - for EU facing systems - the EU AI Act have a very strong focus on model bias. Key takeaways: ✅ Go beyond correlations and understand the causal mechanisms of bias. ✅ See how 'Causality Traces' can map the exact pathway a model takes to reach a decision. ✅ Learn about a real-world case study on an income-prediction model. Read the full article here: https://lnkd.in/eg7YN5-t Matthew Jones Dr Paul Dongha Oriana Medlicott Martin Koder Anuj Agarwal Sophie R. Pavel Nesterov Alok Kulkarni Oli Platt Kris Dickinson Shameek Kundu Jeremy Cogman #AI #MachineLearning #DeepLearning #AIEthics #ResponsibleAI #ModelBias #Fintech #Tech #Innovation #FCA #PRA #EUAIAct #SMCR #ConsumerDuty

Dr Don Liyanage pushing the boundaries of what's possible - again!! This time with some pretty gnarly GPU optimisation which accelerates our tech and has been open-sourced to help the community too :) Struggling with GPU VRAM limits? That frustrating "Out of Memory" error can bring large-scale AI & data science projects to a screeching halt. Standard libraries just aren't built to handle tensors that exceed a GPUs' memory capacity.   We are excited to announce Open Tikos Utilities, a lightweight, open-source Python/CUDA package designed to break through these VRAM barriers. By using a "sliced processing" architecture, it performs massive element-wise tensor operations on consumer-grade hardware without OOM errors.   The results? A staggering 400x performance increase over standard memory safe CPU methods and 2.2x+ faster than a highly optimised, multi-threaded NumPy implementation on large-scale tasks.   Drastically accelerates the generation of Causality Traces (our tech), cutting processing times from minutes down to seconds and making deep model inspection operationally viable.   We are open-sourcing this tool to empower researchers and developers everywhere. Let us make hardware a catalyst for innovation, not a limitation.   Dive deeper and see how it works: 📖 Read about the innovation on our blog: https://lnkd.in/eAm-e6-w   💻 Install the package via PyPI: https://lnkd.in/eAGpDX9J   🔬 Explore the full research paper: https://lnkd.in/eHGKiiqW   #Python #CUDA #GPU #OpenSource #AI #MachineLearning #DataScience #HPC #DeepLearning #VRAM #LLM #NVIDIA

📢📢PRODUCT RELEASE: Finally, a way to understand why LLMs say the things they do. Excellent work by our tech team lead by Dr Don Liyanage. Our latest release tackles the most critical issues with LLMs head-on. 👏👏👏 When you need to build reliable, controllable systems, you have to be able to understand why your models generate particular responses. For those working with open-weights LLMs, getting this insight just became a lot easier. #AIAssurance #TrustworthyAI #LLMs

📢📢NEWS: TIKOS™ Joins Google for Startups AI Program We are thrilled to announce that TIKOS™ has been selected to join the prestigious Google for Startups AI program! This is a significant milestone and further validates our #AIAssurance solutions address a critical market need — providing AI system developers and deployers with the right tools for responsible innovation, deployment and monitoring. Acceptance onto the program is limited to startups working on core AI technologies who have secured early stage VC funding. Joining the Google for Startups AI program will provide us with access to Google's AI experts, a powerful network of fellow innovators and investors, and $350,000 in cloud credits over the next 2 years. "We are incredibly excited to be a part of the Google for Startups AI program," said Dr. Don Liyanage, our Co-founder, Chief Scientist & CTO. "With the risks of AI now in sharper focus, companies will be judged by how responsibly they act. The support from Google allows us to dramatically accelerate the development of our proprietary solutions to meet this exact challenge." The Google for Startups AI program is designed to help startups like ours build and grow. We look forward to leveraging this opportunity to its fullest as we continue to push the boundaries of Trustworthy AI. https://lnkd.in/eDm98E2m #AI #Startups #GoogleforStartups #Innovation #TIKOS™ #AIAssurance #ResponsibleAI #TrustworthyAI

Latest insight from Dr Don Liyanage, our CTO and Chief Scientist

📢News: Tikos joins Future Space - Bristol Deep Tech Hub TIKOS™ are proud to announce our move to Future Space - Bristol, the award-winning innovation centre 🤝 based at University of the West of England. 💼  We join a vibrant community of deep-tech innovators and start-ups, with access to specialist business support, investors, industry-building events, and more. What’s prompted this move? ✅ Our Mission: To be at the forefront of AI Assurance and help establish the South West as a centre for Trustworthy & Responsible AI ✅ Our Team: Having recruited our founding engineering and marketing teams - it's time to forge the company culture and make use of our multiple whiteboards! Futurespace: https://lnkd.in/eCjW7qsv TIKOS™: https://tikos.tech/ #Futurespace #TrustworthyAI #AIAssurance

📢📢 A very warm welcome to our latest intern Ray G.!! Our partnership with Future Space - Bristol continues to be a great help. Their business support team coordinated a fully-funded (marketing) internship, in partnership with the University of the West of England. (A special thanks to Thanh Quan-Nicholls 👍👍) Here is what Ray thinks of his experience so far... 🌟 My internship with Tikos has been very interesting and educational, I've developed my knowledge, understanding and wider skillset in regard to the industry and my confidence has really increased. 🌱 As a product marketing intern I enjoy the range of tasks, a blend of strategic marketing, web design, and other analytical and creative tasks. 🤝 Don and Mike have trusted me to contribute meaningfully to the company. 🎓 I look forward to continuing to learn and contribute to the company as we navigate #AIAssurance together.     #TrustworthyAI #AIAssurance

📢📢NEWS. Tikos has been approved to provide AI Technical Governance Software & Services to the UK Ministry of Defence (MoD) AI is rapidly transforming defence and security (D&S), but its deployment demands the highest levels of ethics and responsibility. Through the Neutral Vendor Framework for Innovation (NVFi), an entirely compliant procurement route, Tikos can now offer our solutions directly to the MoD. This partnership provides a faster, compliant way to bring groundbreaking AI innovation to defence, aligned with Public Procurement Regulations and Defence Security Standards. More information on Constellia’s Neutral Vendor Framework for Innovation: https://lnkd.in/ei7VyAET #AIProcurement #AIAssurance #ExplainableAI #TrustworthyAI

📢📢NEWS. Tikos has been approved by the Crown Commercial Service (CCS) to provide AI Technical Governance Software & Services to UK public bodies and agencies. The company's inclusion as an approved supplier on the 'Artificial Intelligence Dynamic Purchasing System' (AI DPS*) follows an in depth application and review process. The AI DPS is the primary mechanism for public sector buyers to procure AI services from pre-qualified suppliers - a roster of fewer than 500 companies, representing best-in-class providers globally. The UK Government is targeting £20 billion in savings over the next 10 years as a result of adopting AI technologies**. These services must to adhere to #TrustworthyAI as defined by the Government's AI Regulatory Principles***. Tikos is well placed to accelerate the safe deployment of AI across the public sector to help deliver these benefits. *More about the AI DPS 🔗 https://lnkd.in/eEhAzvMn **More about the UK Government's 'AI Opportunities Action Plan' 🔗 https://lnkd.in/d5QQXCWf ***More about the UK Government's AI Regulatory Principles 🔗 https://lnkd.in/gtiGmv5P #AIAssurance #AIGovernance #ExplainableAI #ResponsibleAI #TrustworthyAI