| viXra.org > Artificial Intelligence > 2501 |
 |
 |
| viXra.org > Artificial Intelligence > 2501 |
|
|
[10] viXra:2501.0154 [pdf] submitted on 2025-01-28 20:19:43
Authors: Basit Ali
Comments: 10 pages, written in English, submitted under CC BY-NC 4.0 license.
This paper proposes a hybrid LSTM-Transformer architecture to train a Named Entity Recognition (NER) model on financial data, such as receipts and invoices. These data types are unstructured and come in various formats, making them difficult to process. The proposed model combines the sequential pattern recognition capabilities of LSTM networks with the contextual sensitivity of Transformer self-attention layers, making it well-suited for financial data applications. This study establishes a modular, design-oriented framework, complete with pseudocode and architectural explanations, to serve as a foundation for future empirical testing. This conceptual work aims to set a benchmark in financial data modeling by addressing domain-specific challenges and providing a scalable structure for subsequent validation.
Category: Artificial Intelligence
[9] viXra:2501.0152 [pdf] submitted on 2025-01-29 04:18:00
Authors: Arvind Sundara Rajan, Ravirajan K
Comments: 10 Pages. CC by attribution
This paper attempted to exhibit the application of Artificial Intelligence(AI) in system for optimizing product assortments in a retail environment. By leveraging AI and machine learning (ML) ML algorithms and techniques, the system analyzed consumer data, sales trends, and inventory levels to dynamically adjust product assortments. The approach integrated predictive analytics and decisionsupport frameworks using the advanced AI applications and novel methods in ML employed to improve customer satisfaction and maximize revenue. This paper discussed the detailed methodology and its appropriate algorithms with opted mathematical explanation and real-life benefits in business problem were derived out of this proposed system, along with its potential to transform retail assortment planning.
Category: Artificial Intelligence
[8] viXra:2501.0144 [pdf] submitted on 2025-01-28 00:41:11
Authors: Ravirajan K, Arvind Rajan
Comments: 5 Pages. CC by attribution.
This paper introduces Generative Action Synthesis (GAS), a novel method for imbuing robots withhuman-like emotional expression during task execution. GAS leverages conditional WassersteinGANs (cWGANs) for action generation conditioned on emotional embeddings, guided by expertdemonstrations and refined via Hamiltonian Monte Carlo. A temporal-hierarchical transformer(THT) synthesizes actions while a Von Mises-Fisher mixture model (vMF-MM) resolves ambiguities.The framework also employs stochastic policy gradients for dynamic adjustment based on real-time feedback and task requirements, with Fokker-Planck-Kolmogorov equations ensuring emotionstability. This approach, integrating generative models with reinforcement learning and structuredemotional embedding, enables robots to exhibit a range of emotional behaviors, including anger,humor, and empathy, leading to more natural and adaptable human-robot interactions. Practicalimplications include advanced applications in caregiving, customer service, and other social domains,highlighting its significance in the development of emotionally intelligent robots
Category: Artificial Intelligence
[7] viXra:2501.0141 [pdf] submitted on 2025-01-28 00:49:32
Authors: Ravirajan K, Arvind Sundara Rajan
Comments: 11 Pages.
The integration of biological principles into artificial olfactory systems has led to significant advancements in odor detection and classification. Inspired by the intricate mechanisms of natural olfaction, researchers are developing sophisticated systems that mimic the functionality of biological olfactory pathways. These systems utilize high-density chemoresistive sensor arrays (HCSA) combined with advanced computational techniques, such as FPGA-accelerated glomerular convergence circuits (FGCC) and hierarchical graph neural networks (HGNN). This bioinspired approach enables real-time adaptive responses to volatile organic compounds (VOCs), enhancing the accuracy and efficiency of odor identification.At the core of these innovations is the multiparametric sigmoidal sensor activation (MPSA), which quantifies VOCs by leveraging the diverse responses of sensor arrays. The implementation of lateral inhibition via programmable synaptic crossbars (LIPSC) further refines odor processing by mimicking neural interactions found in biological systems. Additionally, temporal self-organizing maps (TSOM) facilitate dynamic clustering of odor patterns, allowing for a nuanced understanding of complex odor environments.A novel aspect of this research lies in the Grassmannian manifold embedding (GME) of odor profiles, which provides a mathematical framework for representing and analyzing the multidimensional nature of odors. Coupled with Hamiltonian Monte Carlo-optimized feedback (HMC-FB), this system effectively compensates for drift in sensor readings, ensuring consistent performance over time. By bridging the gap between biological inspiration and technological innovation, these artificial olfactory systems are poised to revolutionize applications ranging from environmental monitoring to food safety and healthcare diagnostics.
Category: Artificial Intelligence
[6] viXra:2501.0099 [pdf] submitted on 2025-01-17 21:32:05
Authors: A. A. Alkadrie
Comments: 14 Pages. (Note by viXra Admin: An abstract with < 400 words is required; also please cite and list scientific references)
The idea of Machines can independently solve problems, serve it as solutions, help us in any kind of task completions and so on, is fascinating. In a general term we identified this kind of machine as intellegence machines. Intellegence, as all of us understand it, is strongly related with conciesness. Does this machines have conciesnesses? this is interesting an interesting question that alot of people have in mind about the intellegence machines. Let's explore this together.
Category: Artificial Intelligence
[5] viXra:2501.0079 [pdf] submitted on 2025-01-13 21:31:43
Authors: Tofara Moyo
Comments: 2 Pages.
We present a novel approach to video generation,leveraging compressed hand-drawn representations and latent diffusion models. Our methodology employs a unique two-stage process, wherein a variational auto encoder generates parametersbased on input text, of a generic equation to be graphed into a frame, and a latent diffusion model refines these frames into photorealistic video content. These graphs are designed to looklike hand drawn replicas of the frames in the dataset. By utilizing hand-drawn-like images as a compressed representation, we effectively reduce the dimensionality of the video generation problem, enabling tighter bottleneck architectures and improved efficiency. Our approach demonstrates significant potential forgenerating lenghty ,high-quality, text-conditioned videos, with applications in multimedia creation, robotics, and beyond.
Category: Artificial Intelligence
[4] viXra:2501.0077 [pdf] submitted on 2025-01-12 10:43:59
Authors: Abdullah M. Ahmad
Comments: 29 Pages.
Synthra represents a groundbreaking technological paradigm that harmonizes blockchain and AI technologies, redefining decentralized systems for the modern era. At its core, Synthra introduces an unprecedented integration of AI-driven mechanisms, such as the Proof-of-Veracity consensus, and the Uploaded Contractual Intelligence (UCI) to ensure immutable, ethical, and highly efficient operations. Synthra is designed to address limitations of traditional blockchain systems, achieving zero gas fees, unparalleled security, and a throughput of up to 1 million transactions per second (TPS).Synthra’ s robust architecture incorporates fail-safe mechanisms like the Self-Destruct Swap Chain (SDSC), Forked-Chain Swap Chain (FCSC), and Binomial Walk Swap Chain (BWSC) which safeguards data integrity against potential network compromises through advanced backup and recovery systems. Furthermore, Synthra is envisioned to extend its capabilities to Quantum-Synthra technology, leveraging Quantum Secure Hashing Algorithms (QSHA) and the innovative Qubyte system to ensure resilience against quantum attacks while maintaining operational scalability.This framework paves the way for a new era of decentralized applications, blending AI precision with blockchain transparency and introducing the Uploaded Contractual Intelligence (UCI) as a deterministic executor of ethical principles. Synthra’ s vision is to enable secure, fast, and reliable platforms that revolutionize industries from social networking to finance while laying the foundation for future Temporal communication systems.
Category: Artificial Intelligence
[3] viXra:2501.0051 [pdf] submitted on 2025-01-09 21:04:36
Authors: Clark M. Thomas
Comments: 5 Pages.
AI directs us toward emerging machine technologythat seemingly everybody has encountered, but few truly comprehend. Fake AI essays, fake AI images, fake reviews, and fuzzy app features disguise how quantity is not always superior to quality. Wisdom and intelligence reside inside our seemingly slow brains, still with 100 trillion synaptic connections. Intuitive machine wisdom separate from and equal to human consciousness is possible, but not yet.The ideal synthesis will optimize the best thoughts of machines and humans to preserve our biosphere.
Category: Artificial Intelligence
[2] viXra:2501.0033 [pdf] submitted on 2025-01-07 21:52:23
Authors: Akhil Kumar
Comments: 4 Pages.
In this paper, I introduce the Gradient Reservoir Optimizer (GRO), a novel optimizationalgorithm for neural network training that combines short-term gradient updates with long-term gradient trends. GRO maintains a dynamic "reservoir" of recent gradient directions and utilizes their aggregated trends to influence parameter updates. By blending current gradients with a history-aware reservoir, GRO aims to stabilize convergence and improve robustness to noisygradients. This novel approach provides an additional mechanism to mitigate common issueslike gradient noise and plateaus in training loss. I demonstrate the theoretical underpinnings of GRO, provide its algorithmic structure, and evaluate its performance on benchmark datasets. The results show promise for GRO as a viable alternative to existing optimizers like SGD, Adam, and RMSProp. Additionally, GRO offers flexibility for tuning the influence of historical gradients, making it adaptable across a variety of tasks and architectures.
Category: Artificial Intelligence
[1] viXra:2501.0015 [pdf] submitted on 2025-01-05 22:14:18
Authors: Stephane H. Maes
Comments: 24 Pages. All related details of the projects (and updates) can be found and followed at https://shmaes.wordpress.com/
The pursuit of Artificial General Intelligence (AGI) has been a prominent goal within the field of artificial intelligence. However, this paper argues that current Generative AI Language Models (GenAI LLMs), such as GPT-4 o1, and similar/later LLMs with similar architectures like o3, are fundamentally incapable of achieving AGI. This argument is supported by examining the intrinsic limitations of LLMs, their operational paradigms, and the essential characteristics that define AGI.We discuss a short experiment performed with all the big LLMs, including the latest ones released by the main different AI providers: extracting and producing a list of URL links from a word document. None of the LLMs succeeded, including the latest from OpenAI, Google, Claude or Perplexity. Instead they all get confused, extract only a subset then, when shown how to do it, they hallucinate the links and never produce a complete list. It happens even when shown how to do it. We take this as a counterexample to statements made by many that, by now, end of 2024, GenAI LLMs would, already reach AGI, or be almost there. In fact we argue that AGI is not about to be reached by LLMs any time soon. They will never reach AGI, without changes away from just being LLMS. Claims to the contrary are unrealistic.The paper presents possible direction to reach AGI, and in particulars our views on how to proceed.
Category: Artificial Intelligence
| Third-Party Links: |
|