Toggle light / dark theme

Artificially engineered biological processes, such as perception systems, remain an elusive target for organic electronics experts due to the reliance of human senses on an adaptive network of sensory neurons, which communicate by firing in response to environmental stimuli.

A new collaboration between Northwestern University and Georgia Tech has unlocked new potential for the field by creating a novel high-performance organic electrochemical neuron (OECN) that responds within the frequency range of human neurons. The team also built a complete perception system by designing other organic materials and integrating their engineered neurons with artificial touch receptors and synapses, which enabled real-time tactile signal sensing and processing.

The research, described in a paper in Proceedings of the National Academy of Sciences, could move the needle on intelligent robots and other systems currently stymied by sensing systems that are less powerful than those of a human.

Their method scrambles laser beams into chaotic patterns, making decryption impossible without a trained neural network. This innovation could revolutionize cryptography.

Holograms for Next-Level Encryption

As the demand for digital security grows, researchers have developed a new optical system that uses holograms to encode information, creating a level of encryption that traditional methods cannot penetrate. This advance could pave the way for more secure communication channels, helping to protect sensitive data.

TISR or video SR (VSR) neural network models are designed to leverage temporal neighbor frames to assist the SR of the current frame and are, therefore, expected to achieve better performance than SISR models19 (Supplementary Note 1). Although TISR models have been widely explored in natural image SR to improve video definition, whether such models can be applied to super-resolve biological images (that is, enhancing both sampling rate and optical resolution) has been poorly investigated. Here, we used the total internal reflection fluorescence (TIRF) SIM, grazing incidence (GI) SIM and nonlinear SIM20 modes of our home-built Multi-SIM system to acquire an extensive TISR dataset of five different biological structures: clathrin-coated pits (CCPs), lysosomes, outer mitochondrial membranes (Mitos), microtubules (MTs) and F-actin filaments (Extended Data Fig. 1). For each type of specimen, we generally acquired over 50 sets of raw SIM images with 20 consecutive time points at 2–4 levels of excitation light intensity (Methods). Each set of raw SIM images was averaged out to a diffraction-limited wide-field (WF) image sequence and was used as the network input, while the raw SIM images acquired at the highest excitation level were reconstructed into SR-SIM images as the ground truth (GT) used in network training. In particular, the image acquisition configuration was modified into a special running order where each illumination pattern is applied 2–4 times at escalating excitation light intensity before changing to the next phase or orientation, so as to minimize the motion-induced difference between WF inputs and SR-SIM targets.

To effectively use the temporal continuity of time-lapse data, SOTA TISR neural networks consist of mainly two important components21,22: temporal information propagation and neighbor feature alignment. We selected two popular types of propagation approaches, sliding window (Fig. 1a) and recurrent network (Fig. 1b), and three representative neighbor feature alignment mechanisms, explicit warping using OF15 (Fig. 1c) and implicit alignment by nonlocal attention23,24 (NA; Fig. 1d) or deformable convolution21,25,26 (DC; Fig. 1e), resulting in six combinations in total. For fair comparison, we custom-designed a general TISR network architecture composed of a feature extraction module, a propagation and alignment module and a reconstruction module (Extended Data Fig. 2) and kept the architecture of the feature extraction module and reconstruction module unchanged while only modifying the propagation and alignment module during evaluation (Methods). We then examined the six models on five different data types: linear SIM data of MTs, lysosomes and Mito, three of the most common biological structures in live-cell experiments, nonlinear SIM data of F-actin, which is of the highest structural complexity and upscaling factor in BioTISR, and simulated data of tubular structure with infallible GT references (Supplementary Note 2). As is shown in Fig. 1f, Extended Data Fig. 3 and Supplementary Fig. 2, all models denoised and sharpened the input noisy WF image evidently, among which the model constructed with a recurrent scheme and DC alignment resolved the finest details compared to the GT-SIM image (indicated by white arrows in Fig. 1f). Furthermore, we calculated time-lapse correlation matrices (Fig. 1g) and image fidelity metrics (Fig. 1h–j) (that is, peak SNR (PSNR) and structural similarity (SSIM)) for the output SR images to quantitatively evaluate the temporal consistency and reconstruction fidelity, respectively. According to the evaluation, we found that recurrent network-based propagation (RNP) outperformed sliding window-based propagation (SWP) in both temporal consistency and image fidelity with fewer trainable parameters (Methods) and propagation mechanisms had little effect on the temporal consistency of the reconstructed SR time-lapse data, while the DC-based alignment generally surpassed the other two mechanisms with a similar number of parameters for all types of datasets (Supplementary Fig. 3).

What exists at the core of a black hole? A research team led by Enrico Rinaldi, a physicist at the University of Michigan, has leveraged quantum computing and machine learning to analyze the quantum state of a matrix model, providing new insights into the nature of black holes.

The study builds on the holographic principle, which suggests that the fundamental theories of particle physics and gravity are mathematically equivalent, despite being formulated in different dimensions.

Two prevailing theories describe black holes from different dimensional perspectives. In one framework, gravity operates within the three-dimensional geometry of the black hole. In contrast, particle physics is confined to the two-dimensional surface, resembling a flat disk. This duality highlights a key distinction between the two models while reinforcing their interconnected nature.

Dr. Simon Stringer. Obtained his Ph.D in mathematical state space control theory and has been a Senior Research Fellow at Oxford University for over 27 years. Simon is the director of the Oxford Centre for Theoretical Neuroscience and Artificial Intelligence, which is based within the Oxford University Department of Experimental Psychology. His department covers vision, spatial processing, motor function, language and consciousness — in particular — how the primate visual system learns to make sense of complex natural scenes. Dr. Stringers laboratory houses a team of theoreticians, who are developing computer models of a range of different aspects of brain function. Simon’s lab is investigating the neural and synaptic dynamics that underpin brain function. An important matter here is the The feature-binding problem which concerns how the visual system represents the hierarchical relationships between features. the visual system must represent hierarchical binding relations across the entire visual field at every spatial scale and level in the hierarchy of visual primitives.

We discuss the emergence of self-organised behaviour, complex information processing, invariant sensory representations and hierarchical feature binding which emerges when you build biologically plausible neural networks with temporal spiking dynamics.

00:00:00 Tim Intro.
00:09:31 Show kickoff.
00:14:37 Hierarchical Feature binding and timing of action potentials.
00:30:16 Hebb to Spike-timing-dependent plasticity (STDP)
00:35:27 Encoding of shape primitives.
00:38:50 Is imagination working in the same place in the brain.
00:41:12 Compare to supervised CNNs.
00:45:59 Speech recognition, motor system, learning mazes.
00:49:28 How practical are these spiking NNs.
00:50:19 Why simulate the human brain.
00:52:46 How much computational power do you gain from differential timings.
00:55:08 Adversarial inputs.
00:59:41 Generative / causal component needed?
01:01:46 Modalities of processing i.e. language.
01:03:42 Understanding.
01:04:37 Human hardware.
01:06:19 Roadmap of NNs?
01:10:36 Intepretability methods for these new models.
01:13:03 Won’t GPT just scale and do this anyway?
01:15:51 What about trace learning and transformation learning.
01:18:50 Categories of invariance.
01:19:47 Biological plausibility.

Pod version: https://anchor.fm/machinelearningstrehttps://en.wikipedia.org/wiki/Simon_S / simon-stringer-a3b239b4 “A new approach to solving the feature-binding problem in primate vision” https://royalsocietypublishing.org/do… James B. Isbister, Akihiro Eguchi, Nasir Ahmad, Juan M. Galeazzi, Mark J. Buckley and Simon Stringer Simon’s department is looking for funding, please do get in touch with him if you can facilitate this. #machinelearning #neuroscience.

https://www.neuroscience.ox.ac.uk/res
https://en.wikipedia.org/wiki/Simon_S
/ simon-stringer-a3b239b4.

When astronomers detected the first long-predicted gravitational waves in 2015, it opened a whole new window into the universe. Before that, astronomy depended on observations of light in all its wavelengths.

We also use light to communicate, mostly . Could we use gravitational waves to communicate?

The idea is intriguing, though beyond our capabilities right now. Still, there’s value in exploring the hypothetical, as the future has a way of arriving sooner than we sometimes think.

By crafting an artificial brain-like environment with microscopic nanopillars, researchers have successfully guided neurons to grow in structured networks. This innovation could revolutionize how scientists study neurological conditions by offering a more accurate way to observe brain cell behavior.