Searcharxiv⌕ Search

arXiv subjects

Hongkai Guo

Publications and source records attributed to Hongkai Guo.

3 recordsLinked to original sources

Humanize MLE Agents: Multi-Agent Session Alternation Beyond Single-Agent Limits

Existing LLM-based agent approaches for long-horizon machine learning engineering (MLE) often use context engineering, implemented through custom harnesses that manage tools, skills, and accumulated experience. However, we find that as single-session execution lengthens, the agent's reasoning intensity declines alongside slowing progress, summarized as long-horizon deceleration, and the benefits of these custom harnesses vanish as the underlying model improves, summarized as harness staleness. These findings suggest that their benefits may not persist across versions, which motivates a workflow-level approach in which agents build on one another's work across sessions. We model these alternating agent sessions as a context-reset semi-Markov decision process (CR-SMDP) and establish sufficient conditions for alternation to exceed single-agent reference limits, derive an explicit experiment-budget bound, and characterize when session caps preserve this advantage. Guided by this analysis, we introduce Humanize MLE Agents (HMA), which alternates two agents on a shared workspace, limits each session to at most k experiments and resets the context between sessions, without additional task-specific harness or prior knowledge supplements. Across 75 Kaggle tasks in MLE-bench, HMA reaches a state-of-the-art any-medal rate of 78.2%, exceeding the strongest open-source baseline by 8.0 points with a quarter of its time budget. Moreover, among six configurations spanning five heterogeneous agent pairs, HMA improves any-medal rate over the constituent baselines by 6.4 points on average, demonstrating cross-pair generality. A supplementary case on kernel optimization reaches 878 cycles, within 1.6% of the community leaderboard leader, demonstrating applicability beyond MLE.

cs.AI↗

Inclusive $η'$ production in B decays and the Enhancement due to charged technipions

The new contributions to the charmless B decay $B \to X_{s}η^{\prime}$ from the unit-charged technipions $P^{\pm}$ and $P^{\pm}_{8}$ are estimated. The technipions can provide a large enhancement to the inclusive branching ratio: $Br(B \to X_{s}η^{\prime}) \sim 7\times 10^{-4}$ for $m_{p1}=100GeV$ and $m_{p8}=250 \sim 350 GeV$ when the effect of QCD gluon anomaly is also taken into account. The new physics effect is essential to interpret the CLEO data.

hep-ph↗

The rare K-decays in the Multiscale Walking Technicolor Model

We calculate the one-loop $Z^0$-penguin contributions to the rare K-decays, $K^+ \to π^+ ν\bar ν$, $K_L \to π^0 ν\bar ν$ and $K_L \to μ^+ μ^-$, from the unit-charged technipions $π_1$ and $π_8$ in the framework of the Multiscale Walking Technicolor Model. We find that: (a) the $π_1$ and $π_8$ can provide one to two orders enhancements to the branching ratios of the rare K-decays; (b) by comparing the experimental data of $Br(K^+ \to π^+ ν\bar ν)$ with the theoretical prediction one finds the lower mass bounds, $m_{p8} \geq 249 GeV$ for $F_Q=40GeV$ and $m_{p1}=100 GeV$; (c) by comparing the experimental data of $Br(K_L \to μ^+ μ^-)$ with the theoretical predictions one finds the lower bounds on $m_{p1}$ and $m_{p8}$, $m_{p1} \geq 210 GeV$ if only the contribution from $π_1$ is taken into account, and $m_{p8} \geq 580 GeV$ for $m_{p1}=210 GeV$, assuming $F_Q=40GeV$; (d) the assumed ranges of the masses $m_{p1}$ and $m_{p8}$ in the Multiscale Walking Technicolor Model are excluded by the rare K-decay data.

hep-ph↗