Quantum circuit architecture impacts trainability via Jacobian rank deficiency

Researchers have identified that the architecture shape of quantum neural networks (QNNs) significantly impacts their trainability, even when the total encoding budget remains constant. They found that structural rank deficiency in the Jacobian matrix leads to 'structural gradient starvation,' where an increasing number of parameters become decoupled from the loss function as the parameter count grows. Parallel architectures were shown to avoid this issue, maintaining a non-zero minimum Jacobian eigenvalue. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Identifies a key architectural constraint for training quantum neural networks, potentially guiding future QNN design.

RANK_REASON Academic paper detailing a new finding about QNN trainability. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• QNN

• paper
• other