mHC: Manifold-Constrained Hyper-Connections
This work addresses training and scalability problems for deep learning models using hyper-connections, representing an incremental advancement in neural network architecture design.
The paper tackles the training instability and scalability issues in Hyper-Connections (HC) by proposing Manifold-Constrained Hyper-Connections (mHC), which projects the residual connection space onto a manifold to restore identity mapping and includes infrastructure optimization, resulting in tangible performance improvements and superior scalability.
Recently, studies exemplified by Hyper-Connections (HC) have extended the ubiquitous residual connection paradigm established over the past decade by expanding the residual stream width and diversifying connectivity patterns. While yielding substantial performance gains, this diversification fundamentally compromises the identity mapping property intrinsic to the residual connection, which causes severe training instability and restricted scalability, and additionally incurs notable memory access overhead. To address these challenges, we propose Manifold-Constrained Hyper-Connections (mHC), a general framework that projects the residual connection space of HC onto a specific manifold to restore the identity mapping property, while incorporating rigorous infrastructure optimization to ensure efficiency. Empirical experiments demonstrate that mHC is effective for training at scale, offering tangible performance improvements and superior scalability. We anticipate that mHC, as a flexible and practical extension of HC, will contribute to a deeper understanding of topological architecture design and suggest promising directions for the evolution of foundational models.