New T-KAN architecture boosts high-frequency trading forecasts

Researchers have introduced Temporal Kolmogorov-Arnold Networks (T-KAN), a novel architecture designed to improve high-frequency trading (HFT) forecasting. Unlike traditional models that struggle with noisy, non-linear limit order book data and alpha decay, T-KAN utilizes learnable B-spline activation functions to capture signal shapes rather than just magnitudes. This approach demonstrated a 19.1% relative improvement in F1-score at a k=100 horizon and achieved a 132.48% return compared to DeepLOB's significant drawdown. The T-KAN model also offers enhanced interpretability and is optimized for low-latency FPGA implementation. AI

IMPACT Potential to significantly improve predictive accuracy and profitability in high-frequency trading environments.

RANK_REASON Academic paper introducing a new model architecture. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Ahmad Makinde
• B-spline
• FI-2010 dataset
• field-programmable gate array
• high-frequency trading
• Takayuki Kanaseki
• Temporal Kolmogorov-Arnold Networks

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →