Xiangning Chen

Stabilizing Differentiable Architecture Search via Perturbation-based Regularization

2020 · arXiv (Cornell University)

Differentiable architecture search (DARTS) is a prevailing NAS solution to identify architectures. Based on the continuous relaxation of the architecture space, DARTS learns a differentiable architecture weight and largely reduces the search cost. However, its …

Symbol tuning improves in-context learning in language models

2023 · arXiv (Cornell University)

We present symbol tuning - finetuning language models on in-context input-label pairs where natural language labels (e.g., "positive/negative sentiment") are replaced with arbitrary symbols (e.g., "foo/bar"). Symbol tuning leverages the intuition that when a model …

A Continuous-Field Photonic Chip Architecture: 3D In-Memory Computing with Dynamic Weight Modulation via PHAT B³D-HPA Architecture & PDMM Instruction Set Specification (v3.53 & v1.3)

2026 · Zenodo (CERN European Organization for Nuclear Research)

This record contains the complete foundational documentation for the B³D-HPA (3D Body-High Performance Architecture) photonic computing framework. B³D-HPA Continu3.53.pdf: The primary architectural whitepaper detailing the dual-modality (deterministic/non-deterministic) framework, GPA (Geometric Polarization Arithmetic) logic, and the …

B³D-HPV Lanthanide Quartz Photonic Computing Chip Supplementary Memorandum v3.5503

2026 · Zenodo (CERN European Organization for Nuclear Research)

In the B³D-HPV paradigm, matrix inversion is no longer a high-complexity silicon computation, but a physical collapse of the Hermitian Adjoint operator. By leveraging the geometric nature of polarization, we achieve near-zero latency inversion through …