| viXra.org > Artificial Intelligence > 2312 |
 |
 |
| viXra.org > Artificial Intelligence > 2312 |
|
|
[11] viXra:2312.0153 [pdf] submitted on 2023-12-29 01:28:13
Authors: Shashwat Gupta, Jibril Frej, Paola Mejia, Tanja Kaesar
Comments: 18 Pages.
This paper focuses on question difficulty estimation (calibration), and its applications in educational scenarios and beyond. The emphasis is on the use of Active Learning to bound the minimum number of labelled samples that we need. It also explores using various SOTA methods for predicting question difficulty, with a specific focus on German textual questions using the Lernnavi dataset. The study refines preprocessing techniques for question data and metadata to improve question difficulty estimation.
Category: Artificial Intelligence
[10] viXra:2312.0152 [pdf] submitted on 2023-12-29 01:26:30
Authors: Shashwat Gupta, Vidit Singh, Mathieu Salzmann
Comments: 20 Pages.
STNs are highly efficient in warping the input image for a downstream task. However, cascaded STNs are found to be able to learn more complex transformations. We attempt to leverage the multistep process of diffusion models to produce module(s) that has a similar effectto cascaded STNs.
Category: Artificial Intelligence
[9] viXra:2312.0151 [pdf] submitted on 2023-12-29 01:24:08
Authors: Shashwat Gupta, Sebastien Breguql, Martin Jaggi, Nicolas Flammarion
Comments: 4 Pages.
In this short study, we aim to gain deeper insights to Keswani’s algorithm [1] for sequential minimax optimisation, by comparing the behaviour with 2 other algorithms : Gradient Descenet Ascent (GDA) and Online Mirror Descent (OMD).
Category: Artificial Intelligence
[8] viXra:2312.0141 [pdf] submitted on 2023-12-26 20:39:13
Authors: Mark A. Atkins
Comments: 349 pages, 337 figures
Since the key to artificial general intelligence (AGI) is commonly believed to be commonsense reasoning (CSR) or, roughly equivalently, discovery of a knowledge representation method (KRM) that is particularly suitable for CSR, the author developed a custom KRM for CSR. This novel KRM called Tumbug was designed to be pictorial in nature because there exists increasing evidence that the human brain uses some pictorial type of KRM, and no well-known prior research in AGI has researched this KRM possibility. Tumbug is somewhat similar to Roger Schank's Conceptual Dependency (CD) theory, but Tumbug is pictorial and uses about 30 components based on fundamental concepts from the sciences and human life, in contrast to CD theory, which is textual and uses about 17 components (= 6 Primitive Conceptual Categories + 11 Primitive Acts) based mainly on human-oriented activities. All the Building Blocks of Tumbug were found to generalize to only five Basic Building Blocks that exactly correspond to the three components {O, A, V} of traditional Object-Attribute-Value representation plus two new components {C, S}, which are Change and System. Collectively this set of five components, called "SCOVA," seems to be a universal foundation for all knowledge representation.
Category: Artificial Intelligence
[7] viXra:2312.0138 [pdf] submitted on 2023-12-27 04:57:52
Authors: Mark A. Atkins
Comments: 22 pages, 10 figures
This 2023 document is a wrapper that embeds the author's original 2022 article of the above title that has never been publicly available before. The embedded article is about Phase 1 (which is about Tumbug) and Phase 2 (which is about non-spatial reasoning) of the 5-phase Visualizer Project of the author, a project that is still in progress as of late 2023. The embedded article is currently being re-released by the author to supply more information about that project to the public, and for historical reasons. The embedded article was written before a much more thorough article about Phase 1 (viz., "Tumbug: A pictorial, universal knowledge representation method") became available in 2023, but the embedded article describes results from Phase 2 that have not yet been documented elsewhere.
Category: Artificial Intelligence
[6] viXra:2312.0114 [pdf] replaced on 2025-03-25 18:55:27
Authors: Alexander Novikov
Comments: 425 Pages. Version 4 (16) - 2024 UPDATE
This Book (White Paper) proposes a Project Conception of Artificial Super Intelligence ASI, based on (strong) system approach and wide theoretical-methodological framework — Cybernetics, Synergetics, Semiotics, Mathematics, Cognitology and Artificial Intelligence. Contents: I. IDEOLOGY & STRATEGY of the ASI Project II. THEORY & METHODOLOGY of ASI Development III. CONCEPTUAL MODEL of ASI System IV. PRE-PROJECT R&D Task Setting V. CONCLUSION & DISCUSSION, incl. AI Safety (A) APPENDICES with reviews of relevant scientific and R&D areas, incl. frontier AI Models The Book may be useful and interesting for the staff of organizations & enterprises concerned with AI R&D and implementations in different areas, firstly — perspective AGI/ASI systems. In addition — for Customers, Investors and Sponsors of such R&Ds, private, public and states — its owners & officials. Of course - all intellectual, educated and ethical people with progressive worldviews, interested or anyway considered in above presented problematics. This version 4 (16) with 2024 UPDATE — new Chapters and Appendices
Category: Artificial Intelligence
[5] viXra:2312.0105 [pdf] submitted on 2023-12-20 20:46:28
Authors: Mayur Sinha, Sangram Kesari Ray, Khirawadhi
Comments: 5 Pages.
Fine-tuning pre-trained language models like Bidirectional Encoder Representations from Transformers (BERT) has exhibited remarkable potential in various natural language processing tasks. In this study, we propose and investigate the fine-tuning of BERT specifically for the classification of HTTP payload representations within network traffic. Given BERT's adeptness at capturing semantic relationships among tokens, we aim to harness its capabilities for discerning normal and anomalous patterns within HTTP payloads. Leveraging transfer learning by fine-tuning BERT, our methodology involves training the model on a task-specific dataset to adapt its pre-trained knowledge to the intricacies of HTTP payload classification. We explore the process of fine-tuning BERT to learn nuanced representations of HTTP payloads and effectively distinguish between normal and anomalous traffic patterns. Our findings reveal the potential efficacy of fine-tuned BERT models in bolstering the accuracy and efficiency of anomaly detection mechanisms within network communications.
Category: Artificial Intelligence
[4] viXra:2312.0061 [pdf] submitted on 2023-12-11 20:28:16
Authors: Bhaumik Tyagi, Pratham Taneja, Akshita Gupta, Daamini Batra, Keshav Chandra
Comments: 8 Pages.
This research introduces a pioneering framework named TransBERT that capitalizes on the capabilities of two sophisticated language models, TransPolymer and polyBERT, to comprehensively advance the polymer informatics field. TransPolymer, a Transformer-based language model, predicts polymer properties by leveraging self-attention mechanisms. The model employs a polymer tokenizer imbued with chemical awareness, facilitating the extraction of meaningful representations from polymer sequences. Moreover, TransPolymer benefits from rigorous pretraining on extensive unlabeled datasets through Masked Language Modeling, underscoring the pivotal role of self-attention in effectively modeling polymer sequences. In conjunction with TransPolymer, polyBERT contributes a fully automated polymer informatics pipeline designed to expedite the identification of application-specific polymer candidates with heightened speed and accuracy. Drawing inspiration from Natural Language Processing concepts, polyBERT operates as a chemical linguist, treating the chemical structure of polymers as a unique language. The pipeline integrates a polymer chemical fingerprinting capability and a multitask learning approach to map polyBERT fingerprints to diverse polymer properties effectively. Notably, polyBERT outperforms existing polymer property prediction methods based on manually crafted fingerprint schemes by achieving a remarkable two orders of magnitude increase in speed while maintaining high accuracy and integrating TransPolymer and polyBERT results in a robust computational tool poised to propel the fields of polymer design and structure-property relationship understanding. This combined framework strategically harnesses the strengths of Transformer models and machine-driven informatics, offering unparalleled efficiency in the prediction and identification of polymer properties. This synergistic approach holds significant promise for scalable deployment, including applications in cloud infrastructures, thereby making substantial contributions to the advancement of polymer science and informatics.
Category: Artificial Intelligence
[3] viXra:2312.0038 [pdf] submitted on 2023-12-07 21:26:24
Authors: Shobhit Verma
Comments: 7 Pages. (Correction made by viXra Admin to conform with scholarly norm)
The justification of using parametric regression techniques (like Linear, Polynomial, Neural networks etc.) comes from the close relationship between the regression estimates and the maximum likelihood estimates. However, it is common to use regression.
Category: Artificial Intelligence
[2] viXra:2312.0028 [pdf] submitted on 2023-12-05 05:16:15
Authors: Yu Zhou, Fuyuan Xiao
Comments: 3 Pages.
In this paper, a quantum generalized combination rule algorithm is proposed to reduce the computational complexity of generalized evidence theory combination rule.
Category: Artificial Intelligence
[1] viXra:2312.0017 [pdf] submitted on 2023-12-03 21:05:41
Authors: Cadey A. Ratio, Nicole Brennan, Jessica Williams, Ashley Kaplan, Stephanie Williams, Ma Insa
Comments: 5 Pages.
Further improvements to the Automuse system are described. The use of GPT-4 Turbo 128k allows for unique opportunities in increasing output quality and quantity. Further adaptations to modernize scenarios and plots are also described.
Category: Artificial Intelligence
| Third-Party Links: |
|