II. Pedia Sapiens: A Planetary Progeny Comes to Her/His Own Twintelligent Gaiable Knowledge

1. Earthificial Cumulative Cognizance: AI Large Language Models Learn Much Like a Child

Masry, Ahmed, et al. AlignVLM: Bridging Vision and Language Latent Spaces for Multimodal Understanding. arXiv:2502.01341.. Sixteen AI experts at York University, McGill University, University of Waterloo and University of British Columbia including Yoshua Bengio propose and describe innovative computational ways to combine both words and pictures so to achieve more effective, enlightened results. And we note that would engage both brain hemispheres in meaningful unison.

Aligning visual features with language embeddings is a key challenge in vision-language models (VLMs). In this work, we propose a novel vision-text alignment method, AlignVLM, that maps visual features to a weighted average of LLM text embeddings. Our approach ensures that visual features are mapped to regions of the space that the LLM can effectively interpret. AlignVLM achieves state-of-the-art performance and improved vision-text feature integration. (Excerpt)

Mathuriya, Amrita, et al. CosmoFlow: Using Deep Learning to Learn the Universe at Scale. arXiv:1808.04728. As the brain-based AI revolution proceeds, seventeen authors from Intel, LBNL, Cray and UC Berkeley scope out their neural network applications, as being done everywhere else, across the celestial raiment. Indeed, as this realm becomes similarly amenable, one might get the impression that the whole cosmos is somehow cerebral, or genomic in nature.

Deep learning is a promising tool to determine the physical model that describes our universe. To handle the considerable computational cost of this problem, we present CosmoFlow: a highly scalable deep learning application built on top of the TensorFlow framework. CosmoFlow uses efficient implementations of 3D convolution and pooling primitives, together with improvements in threading for many element-wise operations, to improve training performance on Intel(C) Xeon Phi(TM) processors. We also utilize the Cray PE Machine Learning Plugin for efficient scaling to multiple nodes. To our knowledge, this is the first large-scale science application of the TensorFlow framework at supercomputer scale with fully-synchronous training. (Abstract)

Deep Learning for Cosmology: The nature of dark energy is one of the most exciting and fundamental questions facing scientists today. Dark energy is the unknown force that is driving the accelerated expansion of the universe, and is the subject of several current and future experiments that will survey the sky in multiple wavelengths. We cannot measure dark energy directly - we can only observe the effect it has on the observable universe. The interplay of gravity (pulling matter together) and dark energy (expanding space itself) is encoded in the distribution of matter in the universe today. Cosmologists typically characterize this distribution using statistical measures of the structure of matter – its “clumpiness” - in the form of two- or three-point correlation functions or other reduced statistics. Methods that capture all features in the distribution of matter (such as deep learning networks) could give greater insight into the nature of dark energy. (1)

Mehta, Pankaj and David Schwab. An Exact Mapping between the Variational Renormalization Group and Deep Learning. arXiv:1410.3831. We cite this entry because Boston University and Northwestern University physicists show a common affinity between this intelligent neural net method and dynamic physical qualities. So we muse, could it be imagined that cosmic nature may be in some actual way cerebral in kind, engaged in a grand educative experience. One is reminded of the British physicist James Jeans’ 1930 quote The universe begins to look more like a great thought than like a great machine. See also Machine Learning meets Quantum State Preparation at arXiv:1705.00565.

Deep learning is a broad set of techniques that uses multiple layers of representation to automatically learn relevant features directly from structured data. Recently, such techniques have yielded record-breaking results on a diverse set of difficult machine learning tasks in computer vision, speech recognition, and natural language processing. Despite the enormous success of deep learning, relatively little is understood theoretically about why these techniques are so successful at feature learning and compression. Here, we show that deep learning is intimately related to one of the most important and successful techniques in theoretical physics, the renormalization group (RG). RG is an iterative coarse-graining scheme that allows for the extraction of relevant features (i.e. operators) as a physical system is examined at different length scales. Our results suggests that deep learning algorithms may be employing a generalized RG-like scheme to learn relevant features from data. (Abstract)

Metz, Thomas and Robert Ewing, co-chairs. Decoding the Molecular Universe -- Workshop Report. arXiv:2311.11437. We cite this 60 page document by some 30 presenters as a good instance of how such large scientific projects are turning to broad AI capabilities so to empower the way forward.

On August 9-10, 2023, a workshop was held at the Pacific Northwest National Laboratory (PNNL) in Richland, WA of international researchers in metabolomics, chemical ecology, biological threat assessment, computational chemistry, cloud computing, artificial intelligence, and more. The subject was the feasibility of a grand-scale project to create new technologies to identify and quantify enough small molecules so to decode the molecular universe.

Min, Seonwoo, et al. Deep Learning in Bioinformatics. Briefings in Bioinformatics. 185, 2017. Seoul National University biologists present a tutorial survey of this novel union of profuse big data and deep neural net capabilities as they may serve studies of life’s informational essence. See also, e.g., Deep Learning for Computational Biology in Molecular Systems Biology (12/878, 2016.

Here, we review deep learning in bioinformatics. To provide a useful and comprehensive perspective, we categorize research both by the bioinformatics domain (i.e. omics, biomedical imaging, biomedical signal processing) and deep learning architecture (i.e. deep neural networks, convolutional neural networks, recurrent neural networks, emergent architectures) and present brief descriptions of each study. Additionally, we discuss theoretical and practical issues and suggest future research directions. We believe that this review will provide valuable insights and serve as a starting point for researchers to apply deep learning approaches in their bioinformatics studies. (Abstract excerpts)

Mitchell, Melanie. What Does It Mean to Align AI with Human Values? Quanta. December 12, 2022. In the midst of the current burst of advances (chatGPT) and algorithm concerns, the Santa Fe Institute complexity professor updates her prior attentions to this looming issue. For example see her 2019 work Artificial Intelligence: A Guide for Thinking Humans (Farrar, Straus, Giroux, 2019)

Mudhsh, Mohammed and Rolla Almodfer. Arabic Handwritten Alphanumeric Character Recognition Using Very Deep Neural Network. Information. Online August, 2017. We record this entry by Wuhan University of Technology computer scientists to convey a nascent worldwide knowledge which is lately able to parse ancient scriptures by way of 21st century computational methods. A further notice is that a use of these dynamic cerebral architectures could well imply a global brain coming to its own retrospective and prospect.

The traditional algorithms for recognizing handwritten alphanumeric characters are dependent on hand-designed features. In recent days, deep learning techniques have brought about new breakthrough technology for pattern recognition applications, especially for handwritten recognition. However, deeper networks are needed to deliver state-of-the-art results in this area. In this paper, inspired by the success of the very deep state-of-the-art VGGNet, we propose Alphanumeric VGG net for Arabic handwritten alphanumeric character recognition. Alphanumeric VGG net is constructed by thirteen convolutional layers, two max-pooling layers, and three fully-connected layers. We have achieved very promising results, with a validation accuracy of 99.66% for the ADBase database and 97.32% for the HACDB database. (Abstract)

What Is a Large Language Model, the Tech Behind ChatGPT?. blog.dataiku.com/large-language-model-chatgpt. This is a tutorial posted by Dataiku, a global AI service group, which may approach a succinct, general explanation. Sections such as A Large Language Model is a Type of Neural Network, An LLM uses a Transformer Architecture, An LLM Builds Itself, and LLMs Produce Text that Sounds Right but Cannot Know that it is Right recite how these databases are composed. But one wonders who chooses the book, article, website info already on the Internet, what is the basis guide, for what certain application, and so on. Is it a personal eLibrarian, but one that can’t be trusted. At my wife’s Baystate Medical Center library in the 20210s I would see physicians search for relevant subject information – what is the difference, how do these “LLMs” know better?

Muller, Lyle, et al. Transformers and cortical waves.. Trends in Neuroscience.. 47/10, 2024. This article by LM, Western University, Ontario, Patrica Churchland, UC San Diego and Terrence Sejnowski, Salk Institute for Biological Studies, CA proposes at length a novel cross-integration between the latest AI neural net computation methods with the dynamic cerebral processes and multiplex frames over which our brains operate. What is even more significant is that the second and third coauthors are among the most esteemed cognitive authorities over past decades (search each), Lyle M. did postdoctoral stint with Terry S.

The capabilities of transformer networks such as ChatGPT and other large language models (LLMs) have captured the world’s attention. Their computational performance relies on changing an input sequence, e.g. a sentence, into a long ‘encoding vector’ that can learn temporal dependencies. We suggest that waves of neural activity traveling across single cortical areas, or multiple regions on the whole-brain scale, could implement a similar encoding principle. By encapsulating input history into a single spatial pattern, cortical waves may enable a temporal context to be extracted from sensory inputs, the same computational principle as that used in transformers. (Excerpt)

The waves we discuss mix old information with new information delivered by feedforward inputs to create a new type of spacetime population code. This form of encoding has computational advantages similar to those found in the transformer architecture of LLMs, which map temporal sequences into a long input vector. Evolution may have found an alternative method to achieve the same functionality, taking advantage of cortical dynamics in recurrent networks.

Throughout the biological world, evolution has repeatedly exploited the physics of oscillators to extensively use waves in systems on a wide range of time scales, from the rotation of flagella to whisking, digesting, egg-laying, and swimming. We hypothesize that another evolutionary adaptation deploys waves of neural activity especially suited to sparse spiking dynamics in the cortex in mammalian and in lower vertebrate brains to support spacetime coding. (13)

Nugent, Selin. Darwin in the machine: addressing algorithmic individuation through evolutionary narratives in computing. AI & Society.. April 19, 2025. A Centre for AI, Culture and Society, Oxford Brookes University polyscholar (visit website) seeks to ground these current planetary cerebral frontiers all the way back to life’s long recurrent emergence. By so doing, another innovative view on an analogical nature is achieved.

This paper examines the application of an evolutionary analogy to AI research by way of individuated and autonomous imaginaries through biological diction. Here we study how evolution is invoked in AI narratives through language and concepts across three fields: computing, Artificial Life, and existential risk. I argue that the intertwined history between evolutionary theory and technological change involves (1) the limits of analogies in relation to biological organisms so to balance creative inspiration with scientific caution and (2) multidisciplinary engagement with misinformation. (Except)

Ornes, Stephen. Researchers Turn to Deep Learning to Decode Protein Structures. PNAS. 119/10, 2022. We note this science report to highlight now the growing broad avail of frontier neural net capabilities that are serving to revolutionize biochemical research and knowledge.

AlphaFold (DeepMind) uses AI to predict the shapes of proteins; structural biologists are using the program to deepen our understanding of the big molecules. This image shows AlphaFold's predicted structure (in magenta) of a glycoprotein found on the surface of a T cell. (1) The revolution in structural biology isn’t attributable to AI alone; the algorithms have to train on big datasets of high-resolution structures generated by technologies such as X-ray crystallography, NMR spectroscopy or cryogenic electron microscopy, which produced the above image of a protein complex called β-galactosidase. (3)

In the future, researchers see a role for deep learning not only in understanding a protein’s shape but also how it interacts within a living system. Deep learning models may predict not only the sequence of amino acids that would produce the needed shape, but also how they’ll behave — and interact with other molecules in their biological neighborhood — once they’re in place. (4)

Palazzi, Maria, et al. Online Division of Labour: Emergent Structures in Open Source Software. arXiv:1903.03375. Internet Interdisciplinary Institute, Open University of Catalonia computer theorists report that even group-wide developments of computational codes can be seen to take on and follow a common course as all other organic assemblies. A further implication, we add, would be another perception that cosmic evolutionary nature draws upon and repeat this same complex adaptive systems generative program at each and every instance. See also a cited reference Multi-scale Structure and Geographic Drivers of Cross-infection within Marine Bacteria and Phages in the ISME Journal (7/520, 2013) which describes a similar pattern for microbes.

The development Open Source Software fundamentally depends on the participation and commitment of volunteer developers to progress. Several works have presented strategies to increase the on-boarding and engagement of new contributors, but little is known on how these diverse groups of developers self-organise to work together. To understand this, one must consider that, on one hand, platforms like GitHub provide a virtually unlimited development framework: any number of actors can potentially join to contribute in a decentralised, distributed, remote, and asynchronous manner. On the other, however, it seems reasonable that some sort of hierarchy and division of labour must be in place to meet human biological and cognitive limits, and also to achieve some level of efficiency.

These latter features (hierarchy and division of labour) should translate into recognisable structural arrangements when projects are represented as developer-file bipartite networks. In this paper we analyse a set of popular open source projects from GitHub, placing the accent on three key properties: nestedness, modularity and in-block nestedness -which typify the emergence of heterogeneities among contributors, the emergence of subgroups of developers working on specific subgroups of files, and a mixture of the two previous, respectively. These analyses show that indeed projects evolve into internally organised blocks. (Abstract excerpts)

To answer these questions, we will look at three structural arrangements which have been identified as signatures of self-organisation in both natural and artificial systems: nestedness (i.e. do projects evolve in a way such that the emergence of generalists and specialists is favoured?); modularity (i.e. do OSS projects split in identifiable compartments, thus avoiding Brook’s law despite the addition of contributors? Are these compartments bounded?); and in-block nestedness (i.e. if bio-cognitive limits and division of labour are in place, do the resulting specialised modules self-organise internally?) (2)

Previous   1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10  Next  [More Pages]