Newest papers

Mixture-of-Experts for Cryptocurrency Order Execution: Training Stability, Tail Risk, and Failure Modes

Alexander Ardaiz, Varun Budati, Ali Habibnia · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.03369 · cited by 0 works (OpenAlex) · open-access PDF

Vietnamese

Deep taapsion là tăng cường rủi ro, nhưng các chính sách học tập đặc biệt cho việc thực hiện lệnh có thể thay đổi đáng kể trên các hạt đào tạo, vì vậy sáu lợi ích kiến trúc rõ ràng có thể phản ánh các thành tựu đào tạo thuận lợi thay vì các thuộc tính có thể tái tạo của kiến trúc. Chúng tôi đánh giá vanilla Double Deep Q-Learning (DDQL), K-means-partitioned mixtures của các chuyên gia DDQL ở $K \in \{2, 4, 8\$, và các thông số mạng dày đặc biệt chạy không phù hợp với hiệu suất của $K{=}4$ và $K{=}8$. Chúng tôi đánh giá ngân sách chuyên gia trong gia đình trên 5 phút trung bình tích lũy BTC/USDT giới hạn dữ liệu sách lệnh từ Binance. Không có cấu hình học được cải thiện đáng kể

Machine translation (Meta M2M100); the English original follows.

Abstract

Deep reinforcement-learning policies for order execution can vary substantially across training seeds, so apparent architectural gains may reflect favourable training realisations rather than reproducible properties of the architecture. We evaluate vanilla Double Deep Q-Learning (DDQL), K-means-partitioned mixtures of DDQL experts at $K \in \{2, 4, 8\}$, and dense networks parameter-matched to the $K{=}4$ and $K{=}8$ expert budgets on 5-minute mean-aggregated BTC/USDT limit order book data from Binance. No learned configuration significantly improves mean implementation shortfall over DDQL. Under the reported specification, all have higher mean shortfall than TWAP (0.39 bps) and immediate liquidation (0.21 bps) in an environment whose frictionless replay and terminal-urgency penalty make early liquidation nearly costless; 11/100 vanilla-DDQL runs, versus none in either MoE $K{\geq}4$ arm, converge to a policy that waits until forced liquidation. We then decompose this specification on a device-matched baseline. Annealed exploration alone eliminates observed collapses (12/100 to 0/100; exact McNemar $p{=}4.9{\times}10^{-4}$), matching the elimination under expert partitioning. Combining annealed exploration with the aligned reward restores collapse in 19/30 runs; with all three specification changes, it rises to 48/100. In this environment, expert partitioning is unnecessary to suppress collapse and appears to mask a training-specification failure rather than confer an intrinsic performance benefit. No MoE $K{=}8$ run collapses under any of the six specifications tested. Across-seed dispersion is lowest at $K{=}8$ but non-monotone and not robust to family-wise adjustment, while within-policy tail risk worsens monotonically with $K$. The apparent attribution of the failure mode reverses between 30 and 100 seeds, illustrating the importance of repeated-seed evaluation.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Monotonic function, Training (meteorology), Order (exchange), Limit (mathematics), Computer science, Matching (statistics)

Is this paper real? This page was built from OpenAlex and arXiv records, not generated. Software can verify any reference in one call: GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.03369, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txt

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange