Sergei Solovev · HSE University, Faculty of Computer Science
A classification of Bitcoin integration methods in decentralized financial infrastructure. Published in Russian, vol. 5 no. 2 (2026), pp. 3026–3037, CC BY 4.0, under the HSE University Faculty of Computer Science affiliation printed on the article.
Swapping a 2017-era DeepSORT tracker’s detections and appearance encoder for YOLOv8m and OSNet lifts mean HOTA on six MOT-Challenge videos from 40.17 to 52.63, yet hides a regression on the least-precise clip (35.14 vs 39.86). The cause is reproducible: the modern pipeline silently drops DeepSORT’s detector-confidence and NMS admission gate; restoring it recovers the loss. Framed as an audit, not a new method.
Empirical study showing a sample-size sign reversal in naive RAG for Solidity vulnerability detection: +2.0% Macro-F1 at n=100 flips to -2.7% at n=250 on SolidiFI. Argues for bootstrap confidence intervals in any RAG evaluation.
Autonomous AI agent managing an ERC-4626 yield vault via MCDM scoring (APY/Risk/Cost/Stability). EIP-712 signed decisions, UUPS-upgradeable, 67 unit tests, 76,800+ invariant calls with zero violations. Sepolia deployment included.
Dual-branch BiGRU-CNN for high-frequency LOB mid-price prediction. Weighted Pearson rho_w = 0.266, outperforming LightGBM on 58.3% of sequences. Documents feature sufficiency and the negative ensemble effect on HF data (Wunder Fund RNN challenge).
XGBoost + Optuna on 117,091 Slither-labelled Ethereum contracts achieves F1=0.948 using 65 hand-crafted bytecode features in 15 SWC-mapped semantic categories. Binary classification: recall 0.950, MCC 0.832, PR-AUC 0.990.
EasyOCR with heuristic rules versus the Donut end-to-end Transformer on SROIE 2019, with an error taxonomy under image degradation typical of messenger-grade distortions (compression, rotation, blur).