David Sussillo on Persistence, Luck and the Bonds Between Life and Work

Why It Matters

FORCE learning revives training of recurrent networks, bridging gaps between brain modeling and AI, and unlocking new avenues for scientific discovery and commercial applications.

Key Takeaways

• Larry Abbott mentored Sussillo, leading to FORCE learning breakthrough
• FORCE learning trains chaotic recurrent neural networks effectively
• Breakthrough revived interest in recurrent networks for brain modeling
• Enables testable hypotheses linking artificial and biological neural dynamics
• Simulation-based approach turned chaos into computational resource

Pulse Analysis

At the heart of their collaboration lies FORCE learning, a method that tames chaotic recurrent neural networks (RNNs) by harnessing, rather than suppressing, their intrinsic dynamics. Traditional training algorithms such as back‑propagation fail on RNNs because the networks behave like nonlinear dynamical systems with sensitive dependence on initial conditions. FORCE introduces a strong, stable feedback signal that gradually aligns the network’s chaotic trajectories with a desired output pattern, effectively turning the wild horse of chaos into a disciplined performer. This approach not only achieves precise temporal control but also preserves the rich computational repertoire that chaos provides, opening a new avenue for training deep, recurrent architectures.

The implications of FORCE learning ripple through neuroscience and AI. For neuroscientists, trained RNNs become a sandbox for testing hypotheses about cortical dynamics, memory, and decision‑making. In AI, the technique revives interest in recurrent models that can generate stable, long‑range sequences without exploding gradients, a hurdle that has limited their use in industry. Moreover, its emphasis on leveraging chaos aligns with neuromorphic hardware trends, where stochastic dynamics are seen as a resource. As research builds on this foundation, we can expect more biologically plausible AI systems and deeper insights into brain function.

David Sussillo’s memoir illustrates how mentorship, interdisciplinary curiosity, and a willingness to challenge dogma can spark transformative breakthroughs. The partnership between a seasoned theorist and an eager PhD candidate turned a seemingly impossible problem—training chaotic recurrent networks—into a practical solution that reshapes both computational neuroscience and machine learning. By converting chaos into a computational asset, FORCE learning not only advances our understanding of brain-like processing but also offers a powerful tool for engineers building next‑generation AI systems that require robust, flexible temporal dynamics.