arXiv subjects

Yijun Ge

Publications and source records attributed to Yijun Ge.

4 recordsLinked to original sources

arXiv2026

Projecting BrowseComp-Plus onto ClimbMix: Toward More Realistic Corpora for Agentic Search

The BrowseComp-Plus benchmark disentangled the evaluation of agentic search by replacing opaque web search with a fixed corpus, so that an agent's role can be separated from the retriever's. That corpus, however, holds only about 100K documents and was assembled from the supporting documents of the benchmark's own queries plus mined hard negatives, so the evidence and the distractors were both selected per query. We introduce $\text{BrowseComp-Plus}_{\text{CM}}$, which keeps the BrowseComp-Plus questions but relocates their evidence to ClimbMix, a 400B-token, 553M-document mixture of web text released by NVIDIA for pre-training language models and built without reference to any benchmark. Our main contribution is the projection pipeline that makes this possible: it decomposes each question into atomic reasoning hops and grounds every hop in the new corpus, retaining a question only when automatic verification, an independent agent, and human review all confirm that every hop is supported. The pipeline is dataset-agnostic and applies to any benchmark whose questions decompose into verifiable facts. Applied to the 830 BrowseComp-Plus test questions, our pipeline yields 57 fully grounded questions with question-level relevance judgments. Projection shifts the difficulty onto retrieval, as the strongest agent we evaluate loses five points of answer accuracy but sees its evidence recall fall from 84.3% to 21.4% while issuing 63% more search calls. As the first of a series of projections, we release the pipeline, the benchmark, and our analyses at https://github.com/castorini/cmass.

cs.IR↗

arXiv2025

Lighting the Way for BRIGHT: Reproducible Baselines with Anserini, Pyserini, and RankLLM

Retrieval benchmarks for large language models (LLMs) should reflect the long, reasoning-intensive queries typical of retrieval-augmented generation (RAG). We present a systematic study of BRIGHT, a reasoning-focused retrieval benchmark, along with strong, reproducible reference methods integrated into Anserini, Pyserini, and RankLLM. We evaluate lexical, sparse, dense, and fusion-based retrievers, as well as LLM rerankers, under long-query settings. In reproducing BRIGHT's lexical baseline, we identify a key under-documented detail: query-side BM25 (BM25Q), which applies BM25 weighting to the query itself. On long, multi-sentence queries, BM25Q consistently outperforms standard BM25, making it the strongest lexical baseline for reasoning-oriented retrieval. We further audit the BRIGHT corpus, uncovering data quality issues that impact evaluation, and offer mitigation. Finally, we study the generalizability of BM25Q across five additional benchmarks, finding its gains largely specific to BRIGHT, while fusion with standard BM25 provides the most consistent improvements across datasets.

cs.IR↗

arXiv2020

Photoconductivity calculations of bilayer graphene from first principles and deformation-potential approach

We report first-principles calculations of electron-phonon coupling in bilayer graphene and the corresponding contribution to carrier scattering. At the phonon $Γ$ point, electrons with energies less than 200 meV are scattered predominantly by LA$^\prime$ and TA$^\prime$ modes while higher-energy electron scattering is dominated by optical phonon modes. Based on a two-temperature model, heat transfer from electrons with an initial temperature of 2000 K to the lattice (phonons) with an initial temperature of 300 K is computed, and in the overall relaxation process, most of this energy scatters into K-point phonon optical modes due to their strong coupling with electrons and their high energies. A Drude model is used to calculate photoconductivity for bilayer graphene with different doping levels. Good agreement with prior experimental trends for both the real and imaginary components of photoconductivity confirms the model's applicability. The effects of doping levels and electron-phonon scattering on photoconductiviy are analyzed. We also extract acoustic and optical deformation potentials from average scattering rates obtained from density functional theory (DFT) calculations and compare associated photoconductivity calculations with DFT results. The comparison indicates that momentum-dependent electron-phonon scattering potentials are required to provide accurate predictions.

cond-mat.mes-hall↗

arXiv2020

Atomistic Simulation of Phonon and Magnon Thermal Transport across the Ferro-Paramagnetic Transition

A temperature-dependent approach involving Green-Kubo equilibrium atomic and spin dynamics (GKEASD) is reported to assess phonon and magnon thermal transport processes accounting for phonon-magnon interactions. Using body-center cubic (BCC) iron as a case study, GKEASD successfully reproduces its characteristic temperature-dependent spiral and lattice thermal conductivities. The non-electronic thermal conductivity, i.e., the sum of phonon and magnon thermal conductivities, calculated using GKEASD for BCC Fe agrees well with experimental measurements. Spectral energy analysis reveals that high-frequency phonon-magnon scattering rates are one order of magnitude larger than those at low frequencies due to energy scattering conservation rules and high densities of states. Higher temperatures further accentuate this phenomenon. This new framework fills existing gaps in simulating thermal transport across the ferro- to para-magnetic transition. Future application of this methodology to phonon- and magnon-dominant insulators and semiconductors will enhance understanding of emerging thermoelectric, spin caloritronic and superconducting materials.

cond-mat.mes-hall↗