Computational neuroscience
Branch of neuroscience
Follow Computational neuroscience on Notably News to receive short updates to your email — rarely!
We include updates on Artificial intelligence, Artificial general intelligence, Convolutional neural network, Softmax function, Action potential, Effective accelerationism, Artificial consciousness, Free energy principle, Neural oscillation, Hodgkin–Huxley model, AI alignment, Spiking neural network, Organoid intelligence, Integrated information theory, Conference on Neural Information Processing Systems, History of artificial neural networks ... and more.
| March 10 2025 |
Integrated information theory
Nature Neuroscience published a follow-up to the original criticism letter, featuring a rebuttal and a commentary proposing a 'weak IIT' modification to address existing critiques.
|
| 2024 |
Hallucination
David Baker received the Nobel Prize in Chemistry for his work using AI hallucinations to design over ten million new proteins, leading to approximately 100 patents and the founding of more than 20 biotech companies.
|
| July 2024 |
Hallucination
A White House report on fostering public trust in AI research mentions hallucinations in the context of reducing them
|
| 2023 |
Hallucination
Researchers estimated that AI chatbots hallucinate up to 27% of the time, with factual errors present in 46% of generated texts, highlighting the significant reliability challenges in large language models.
|
| 2023 |
Hallucination
Nvidia launched Guardrails, a tool that can be configured to hard-code certain AI responses, providing a method to mitigate hallucination risks in language models.
|
| 2023 |
Hallucination
The Cambridge dictionary updated its definition of hallucination to include the new meaning specific to the field of AI.
|
| September 15 2023 |
Integrated information theory
A letter signed by 124 scholars was published on PsyArXiv, asserting that Integrated Information Theory should be labeled pseudoscience until it becomes empirically testable.
|
| June 2023 |
Dehaene–Changeux model
Initial results of the collaborative research were revealed, showing that none of the Dehaene–Changeux model's predictions passed the pre-registered threshold, while two out of three predictions from integrated information theory were successful.
|
| June 2023 |
Integrated information theory
Initial results of the Templeton Foundation-funded study were revealed, showing that none of GNWT's predictions passed the pre-agreed threshold, while two out of three of IIT's predictions passed.
|
| June 2023 |
Hallucination
Mark Walters sued OpenAI in a Georgia state court after ChatGPT allegedly generated a defamatory mischaracterization of a legal complaint.
|
| June 23 2023 |
Hallucination
Judge P. Kevin Castel dismissed the Mata case and issued a $5,000 fine to Schwartz and another lawyer for submitting fictitious legal precedents.
|
| May 2023 |
Hallucination
OpenAI characterizes hallucinations as 'a tendency to invent facts in moments of uncertainty' and 'a model's logical mistakes'
|
| May 2023 |
Hallucination
Stephen Schwartz submitted six fake case precedents generated by ChatGPT in a legal brief for the Mata v. Avianca personal injury case.
|
| May 2023 |
Hallucination
CNBC describes AI hallucinations as 'fabricating information entirely, but behaving as if spouting facts'
|
| February 2023 |
Hallucination
The Verge defines AI hallucination as 'making up information'
|
| 2022 |
Hallucination
The New York Times published concerns about large language model bots, highlighting potential issues with user confidence in AI-generated output and the risk of hallucinations as AI adoption grew.
|
| November 30 2022 |
Hallucination
OpenAI released ChatGPT in beta version to the public, based on the GPT-3.5 foundation model.
|
| November 17 2022 |
Hallucination
Meta withdrew Galactica due to offensiveness and inaccuracy after its problematic content generation was highlighted.
|
| November 15 2022 |
Hallucination
Meta AI researchers published Galactica, an AI model designed to store, combine, and reason about scientific knowledge, with a warning that its outputs may be unreliable due to potential text hallucinations.
|
| 2021 |
Integrated information theory
Angus Leung and colleagues published a direct application of Integrated Information Theory's mathematical formalism to neural data, focusing on neuronal population activity in flies and demonstrating how Φ^Max can be computed for smaller neural datasets.
|
| July 2021 |
Hallucination
Meta warned during the release of BlenderBot 2 that the system is prone to 'hallucinations', defined as 'confident statements that are not true'.
|
| 2019 |
Dehaene–Changeux model
The Templeton Foundation announced funding over $6,000,000 for an adversarial collaboration to empirically test the Dehaene–Changeux model against integrated information theory, with both theory originators agreeing to experimental protocols and predefined success criteria.
|
| 2019 |
Integrated information theory
New Scientist published an article investigating how consciousness works, presenting radical theoretical answers.
|
| 2019 |
Integrated information theory
Christof Koch published 'The Feeling of Life Itself: Why Consciousness is Widespread but Can't Be Computed', a book discussing integrated information theory.
|
| 2019 |
Integrated information theory
The Templeton Foundation announced funding exceeding $6,000,000 to test empirical predictions of Integrated Information Theory (IIT) against Global Neuronal Workspace Theory (GNWT), with both theory originators agreeing on experimental protocols and conditions for validating their theories.
|
| 2018 |
Cerebellar model articulation controller
A deep CMAC (DCMAC) framework was proposed with a derived backpropagation algorithm for estimating its parameters. Experimental results demonstrated improved noise cancellation performance compared to conventional single-layer CMAC in an adaptive noise cancellation task.
|
| 2018 |
Hallucination
The term 'hallucinations' was used in computer vision to describe instances where non-existent objects are erroneously detected because of adversarial attacks.
|
| 2018 |
Hallucination
Wired magazine reported on potential AI hallucination vulnerabilities in consumer devices and automated systems, highlighting examples of adversarial attacks that could manipulate AI perception, such as an invisible stop sign and misclassified images.
|
| 2017 |
Integrated information theory
Nautilus magazine explored the philosophical question 'Is Matter Conscious?', further examining the implications of Integrated Information Theory.
|
| 2017 |
Hallucination
Google researchers used the term 'hallucination' to describe responses generated by neural machine translation (NMT) models when they are not related to the source text.
|
| 2016 |
Integrated information theory
Aeon magazine published an article discussing the concept of 'consciousness creep'.
|
| 2015 |
Convolutional neural network
Atomwise introduced AtomNet, the first deep learning neural network for structure-based drug design, which uses convolutional neural networks to train on 3-dimensional representations of chemical interactions and discover chemical features.
|
| 2015 |
Convolutional neural network
AtomNet was used to predict novel candidate biomolecules for disease treatments, specifically targeting the Ebola virus and multiple sclerosis.
|
| 2015 |
Convolutional neural network
A multi-layered Convolutional Neural Network was developed that could detect faces from multiple angles, including upside down and partially occluded. The network was trained on 200,000 face images and 20 million non-face images, using batches of 128 images over 50,000 iterations.
|
| February 2015 |
Convolutional neural network
Research paper on Deep Q-networks was published, detailing their application to Atari 2600 gaming and showcasing the potential of using convolutional neural networks in reinforcement learning.
|
| 2014 |
Integrated information theory
MIT Technology Review discussed the challenges and requirements for computer consciousness.
|
| 2014 |
Convolutional neural network
A large 12-layer convolutional neural network was developed that could correctly predict professional Go moves with 55% accuracy, matching the performance of a 6 dan human player.
|
| 2014 |
Convolutional neural network
Dropout regularization method was introduced to reduce overfitting in neural networks, specifically addressing the problem of fully connected layers being prone to overfitting by randomly dropping out nodes during training.
|
| 2014 |
Convolutional neural network
A trained convolutional network, when used directly to play Go without search algorithms, beat the traditional search program GNU Go in 97% of games and matched the performance of Fuego using Monte Carlo tree search with ten thousand playouts per move.
|
| 2014 |
Convolutional neural network
Preliminary results of Deep Q-networks (DQN) were first presented, demonstrating a novel approach combining deep neural networks with Q-learning for reinforcement learning using high-dimensional sensory inputs.
|
| 2014 |
Convolutional neural network
GoogLeNet won the ImageNet Large Scale Visual Recognition Challenge (ILSVRC), increasing mean average precision of object detection to 0.439329 and reducing classification error to 0.06656, demonstrating near-human performance in object recognition.
|
| December 2014 |
Convolutional neural network
Clark and Storkey published a research paper demonstrating that a CNN trained through supervised learning from professional Go game databases could outperform GNU Go and win games against Monte Carlo tree search Fuego 1.1 in significantly less time.
|
| 2013 |
Integrated information theory
Wired magazine published an article featuring a neuroscientist's radical theory about how networks become conscious. The New Yorker also explored the theory, questioning the consciousness of technological devices like iPhones.
|
| 2013 |
Convolutional neural network
Stochastic pooling technique introduced as an alternative to conventional deterministic pooling operations in neural network architectures. The method randomly selects activations within pooling regions according to a multinomial distribution, providing a new approach to regularization in convolutional neural networks.
|
| 2012 |
Integrated information theory
Giulio Tononi published 'Phi: A Voyage from the Brain to the Soul', a book exploring the integrated information theory of consciousness.
|
| 2012 |
Artificial intelligence
Deep learning begins to dominate industry benchmarks and is widely adopted across the AI field.
|
| 2012 |
Convolutional neural network
AlexNet won the ImageNet Large Scale Visual Recognition Challenge, achieving significant performance in image classification.
|
| 2012 |
Convolutional neural network
AlexNet, a GPU-based CNN by Alex Krizhevsky et al., won the ImageNet Large Scale Visual Recognition Challenge, which was a catalytic event for the AI boom.
|
|
We are only showing the most recent entries for this topic. |
|
This contents of the box above is based on material from the Wikipedia articles Hallucination (artificial intelligence), Pooling layer, Cerebellar model articulation controller, Artificial intelligence, Integrated information theory, Convolutional neural network & Dehaene–Changeux model, which are released under the Creative Commons Attribution-ShareAlike 4.0 International License.