Q2B26 Copenhagen | Pekka Pursula, Head of Partnerships, Qutwo

Q2B26 Copenhagen | Pekka Pursula, Head of Partnerships, Qutwo

by Pekka Pursula · View the channel on PSI

Applied Sciences & Engineering Economics & Finance KJBusiness and ManagementKJDBusiness innovation
🎙 Pekka Pursula 👥 6K 📅 October 12, 2026 ⏱ 18 min 👁 56 📄 expert opinion 🧭 2026-10-15
Available in: English (current) Français

Keywords

quantum computingAIQPUbarren plateaushybrid models

Summary

Pekka Pursula, Head of Partnerships at Qutwo, presents an overview of how quantum computing can integrate into the AI workflow. He begins by contextualizing the evolution of AI with compute availability, noting that GPUs power the current wave, but quantum computing is poised to be the next paradigm. The talk outlines a simplified AI workflow: data curation, training, optional compression, and inference. Quantum computing can potentially accelerate training by addressing the data bottleneck and helping with barren plateaus (vanishing gradients). In compression, quantum-inspired methods like tensor train models can reduce model memory usage by 80-90%. For inference, adding a quantum layer to classical neural networks has shown improved accuracy in tasks like medical image recognition and even in small LLMs, though inference latencies are too high for real-time applications. The speaker emphasizes that quantum computing is not yet suitable for frontier LLMs but is beneficial for focused models. He argues that Europe has a strategic advantage in quantum technologies and suggests that wide, shallow circuits might be more relevant than deep, narrow fault-tolerant circuits for AI. The presentation concludes with a call for close integration with classical hardware to reduce latency and improve efficiency.

194 words

Critical Evaluation

Value of the Information & Strength of the Argument

The talk provides a balanced perspective on the potential of quantum computing in AI, avoiding hype by clearly stating where quantum offers advantages and where it does not. The argumentation is structured around concrete examples from literature, such as the data bottleneck being an engineering rather than fundamental problem, and the use of tensor train methods as a quantum-inspired compression technique. The speaker also discusses hybrid models and the importance of latency, which is a critical but often overlooked aspect. However, the presentation remains at an overview level, lacking in-depth technical details or rigorous comparative analysis. The argument that Europe can lead in this domain is plausible but not substantiated with data.

Scientific Rigor, Source Quality, Title Accuracy

The speaker cites several theoretical papers and results without giving explicit citations, which weakens the ability to verify claims. The Q2B conference link in the description serves as the only direct reference. The title of the video is generic and does not convey the content, making it somewhat misleading for viewers seeking specific information. The presentation is consistent with the title in that it discusses quantum computing and AI, but the title does not mention the specific angle. Overall, the scientific rigor is moderate; the talk is more of a strategic overview than a detailed technical review.

224 words

Title / Content Match

The title indicates a conference talk, but does not describe the content specifically. The content matches the expected topic of quantum computing and AI.

Quality & Reliability

6/10

The talk provides a high-level overview of quantum computing applications in AI, referencing several theoretical results but without providing detailed methodologies or citations. The speaker is from a quantum AI startup, which may introduce bias.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

Contribution & Novelties

The talk offers a practical perspective on where quantum computing can plug into AI workflows, highlighting specific bottlenecks and opportunities. It emphasizes that quantum is not a universal solution but can be beneficial in certain applications, such as offline medical imaging and focused optimization problems. The concept of using wide and shallow quantum circuits for AI, as opposed to fault-tolerant deep circuits, is an original contribution to the discussion.

Pour aller plus loin :

  • Quantum machine learning — Overview of methods where quantum computing meets ML, relevant to the hybrid approaches discussed.
  • Tensor train — The tensor train decomposition is a quantum-inspired method for model compression, as mentioned in the talk.
  • Quantum annealing — Related to the use of annealers for optimization problems in small models, referenced in the talk.

130 words

Radar Profile

The profile shows moderate scores across all dimensions, indicating a balanced but not exceptionally deep presentation. The technical level is appropriate for a general technical audience, but the limited source citations and lack of detailed evidence reduce overall reliability.

Reliability 6/10