Neurospace- JHU

Inspiration

We were inspired by classic space retro games.

What it does

Neurospace is a fast-paced signal encoding game where players input combinations of directional controls that are decoded in real time. Using an n=5 input space, the system maps user actions into a compact symbol set and uses a CNN-based decoder to interpret noisy inputs. The goal is to maximize information throughput (bit rate) by balancing speed and accuracy under uncertainty.

How we built it

Designed a discrete input system (n=5) to simulate constrained communication channels Built a CNN decoder to classify input patterns robustly under noise Implemented a real-time feedback loop to measure accuracy and bit rate Modeled performance using an information-theoretic objective (log₂-based bitrate formulation)

Challenges we ran into

Accuracy vs. speed tradeoff: Increasing input complexity improved theoretical bit rate but hurt real-world performance Decoder generalization: The CNN needed to remain stable across noisy, inconsistent user inputs Human factors: Designing a system that is both learnable and high-throughput was non-trivial

Accomplishments that we're proud of

Achieved ~97% accuracy on the easy (n=5) decoding task Successfully implemented a CNN-based real-time decoder Built a working system that directly optimizes for bit rate, not just gameplay Created a novel framework that bridges games + information theory + neural decoding

What we learned

Maximizing bit rate is not just about increasing input space—it requires careful co-design of human interaction and model decoding Simpler input systems can outperform complex ones when considering real-world noise Human usability constraints are just as important as algorithmic performance

What's next for Neurospace

Expand beyond n=5 to adaptive input systems with dynamic symbol sets Improve the CNN with temporal modeling (e.g., sequence-based decoding) Explore applications in brain-computer interfaces and assistive communication Develop a fully polished game with progression, training modes, and competitive play

Built With

• synapse