| ai.viXra.org > Digital Signal Processing > 2508 |
|
 |
| ai.viXra.org > Digital Signal Processing > 2508 |
|
|
[2] ai.viXra.org:2508.0045 [pdf] submitted on 2025-08-14 19:32:10
Authors: Jaba Tkemaladze
Comments: 30 Pages.
This article presents an innovative predictive model of the world based on dynamic updating and adaptive filtering of predicates. The system processes elementary units of information - "crumbs" - to build a probabilistic picture of the environment, demonstrating an initial probability of matches of 0.5 and exponential decay to 0.00001 as the number of counters increases. Key mechanisms include: (1) updating significant patterns with PredictIncrement=2, (2) filtering rarely used predicates while maintaining plasticity balance (γ≥0.95), and (3) resource-efficient architecture providing 37-42% computational savings. Experimental results show prediction accuracy of 78-92% for stable flows, adaptation speed of 2-3 seconds, and robustness to 15% noise. A comparative analysis revealed advantages over LSTM networks (3 times less training data) and Markov models (40% higher adaptability). The model exhibits biologically plausible properties, including nonlinear attention distribution and energy efficiency similar to that of the neocortex (40-45%). Application prospects include IoT, cybersecurity and power system management, and further research is aimed at integrating the temporal model and hierarchical organization of patterns.
Category: Digital Signal Processing
[1] ai.viXra.org:2508.0008 [pdf] submitted on 2025-08-03 21:02:58
Authors: Rickesh Thandalai Natarajan, Surender Thandalai Natarajan
Comments: 7 Pages.
We present a novel algorithmic framework for detecting bias in automated lending systems using large-scale mortgage application data. Our approach employs stratified matching algorithms and statistical hypothesis testing to identify systematic discrimination patterns in financial decision-making systems. Applied to 947,927Home Mortgage Disclosure Act (HMDA) records from2007-2016, our framework detects significant algorithmic bias affecting minority applicants, with Black applicants experiencing 21.1 percentage point lower approval rates than equivalent White applicants. The system achieves96% statistical significance across income-loan amount strata, demonstrating the effectiveness of our bias detection methodology. Our contributions include: (1) a scalable bias detection algorithm for high-volume financial data, (2) robust statistical validation framework for discrimination detection, and (3) empirical evidence of systematic bias in real-world lending algorithms. The framework is generalizable to other algorithmic decision-making domains where fairness is critical.Keywords: Algorithmic bias, fairness in machinelearning, automated decision systems, bias detection, financial technology, mortgage lending.
Category: Digital Signal Processing