arXiv subjects

Zhihua Han

Publications and source records attributed to Zhihua Han.

3 recordsLinked to original sources

arXiv2025

RecIS: Sparse to Dense, A Unified Training Framework for Recommendation Models

In this paper, we propose RecIS, a unified Sparse-Dense training framework designed to achieve two primary goals: 1. Unified Framework To create a Unified sparse-dense training framework based on the PyTorch ecosystem that meets the training needs of industrial-grade recommendation models that integrated with large models. 2.System Optimization To optimize the sparse component, offering superior efficiency over the TensorFlow-based recommendation models. The dense component, meanwhile, leverages existing optimization technologies within the PyTorch ecosystem. Currently, RecIS is being used in Alibaba for numerous large-model enhanced recommendation training tasks, and some traditional sparse models have also begun training in it.

cs.IR↗

arXiv2025

Downloading many-qubit entanglement from continuous-variable cluster states

Many-body entanglement is an essential resource for many quantum technologies, but its scalable generation has been challenging on qubit platforms. However, the generation of continuous-variable (CV) entanglement can be extremely efficient, but its utility is rather limited. In this work, we propose a scheme to combine the best of both qubit and CV approaches: a systematic method to download useful many-qubit entanglement from the efficiently generated CV cluster states. Our protocol is based on one-bit teleportation of the qubit correlation encoded in the displaced Gottesman-Kitaev-Preskill basis. To characterize the practical performance of our scheme, we develop an equivalent circuit to map dominant CV errors to single-qubit preparation errors. Particularly, we relate finite squeezing error to qubit erasure, and show that only 5.4 dB squeezing is sufficient to implement robust qubit memory or quantum computation (QC), and 11.9 dB for fault-tolerant QC. Our protocol can be implemented with the operations that are common in many bosonic platforms.

quant-ph↗

arXiv2024

Linear-optical quantum computation with arbitrary error-correcting codes

High-rate quantum error correcting codes mitigate the imposing scale of fault-tolerant quantum computers but require efficient generation of non-local, many-body entanglement. We provide a linear-optical architecture with these properties, compatible with arbitrary codes and Gottesman-Kitaev-Preskill qubits on generic lattices, and featuring a natural way to leverage physical noise bias. Simulations of hyperbolic surface codes and bivariate bicycle codes, promising families of quantum low-density parity-check codes, reveal a threshold comparable to the 2D surface code with substantially better encoding rates.

quant-ph↗