谛听Diting Lab

Crypto & Blockchain · Tue, Oct 20, 2026

Bitcoin, Ethereum, stablecoins, DeFi, smart contracts, scaling and consensus.

arXiv q-fin.PM

Market Regime-Based Multi-Asset Portfolio Optimization Using Nested Clustered Optimization for Tactical Asset Allocation

In briefRegime-aware multi-asset portfolio allocation using hidden Markov models and nested clustered optimization, tested on Indonesian assets.

SummaryThe paper proposes a tactical asset allocation framework that combines Gaussian Hidden Markov Model regime detection with Nested Clustered Optimization, with regimes identified out of sample. It is tested on an Indonesian multi-asset universe including government bonds, LQ45 equities, gold, CNY/IDR exchange rates, and Bitcoin over 56 out-of-sample months from 2021 to 2025. The regime-aware version limits maximum drawdown to 4.44%, but the regime-agnostic NCO shows a lower drawdown of 2.19% and higher cumulative return. The authors find that a rolling 24-month HMM switches regimes in 55 of 56 months, causing regime drift that erodes the expected gains.

Supriadi, Bambang Siswoyo

arXiv cs.CR · cs.SE

When EVM Defaults Break: Three SDK Patterns for UTXO-Family Wallets

In briefWhy Ethereum-style SDK defaults fail on UTXO-style chains: three reusable design patterns

SummaryThis paper argues that mainstream Web3 SDKs such as ethers.js and viem assume an account-based ledger, a single secp256k1 signer, and transactions built without input selection. These assumptions break on UTXO-family chains like Nervos CKB's Cell model and when wallets use different cryptographic primitives. Using the TypeScript SDK CCC, the authors extract three recurring patterns from its release history and exported APIs: a declarative transaction composer, a unified signer abstraction, and a drop-in connector UI component. The paper presents these as transferable design lessons for SDKs serving UTXO-family ledgers and heterogeneous wallets.

Hanzheng Zhang, Hongzhou Chen

arXiv cs.CV · cs.AI

Mise-en-Video: Fine-Grained 3D Control for Multi-Shot Video Generation

In briefMise-en-Video: Fine-Grained 3D Control for Multi-Shot Video Generation

SummaryThis paper presents Mise-en-Video (MeV), a framework that lets creators stage multiple shots in 3D space and then generate the matching multi-shot video. It combines a hierarchical attention mechanism for narrative consistency across levels with 3D-aware layout guidance for precise per-shot spatial control. The authors also built a 160K-video dataset with hierarchical multi-shot texts and oriented 3D bounding boxes for training and evaluation. Experiments report that MeV outperforms strong baselines in video quality, inter-shot consistency, and subject-level controllability.

Yingshuang Zou, Xiaoyu Shi, Yawen Luo et al.

arXiv cs.CR

Proof-of-Machine-Learning-Inference: Mining Knowledge, Not Hashes

In briefA blockchain consensus method where miners run AI inference instead of hash puzzles

SummaryThis paper proposes Proof-of-ML-Inference (PoML), a consensus protocol in which miners perform user-requested machine-learning inference instead of computing hash puzzles. The authors formalize the computational assumptions needed for ML inference to replace hash work and argue that LLMs and diffusion models satisfy them. They analyse security by reducing it to the Bitcoin Backbone model and evaluate block-time stability in a simulated environment.

Wan Ki Wong, Michele Ciampi, Rik Sarkar