Lifeboat Foundation

Filmmakers may soon be able to stabilize shaky video, change viewpoints and create freeze-frame, zoom and slow-motion effects – without shooting any new footage – thanks to an algorithm developed by researchers at Cornell University and Google Research.

The software, called DynIBar, synthesizes new views using pixel information from the original video, and even works with moving objects and unstable camerawork. The work is a major advance over previous efforts, which yielded only a few seconds of video, and often rendered moving subjects as blurry or glitchy.

The code for this research effort is freely available, though the project is at an early stage and not yet integrated into commercial video editing tools.

Generative AI techniques like ChatGPT, DALL-e and Codex can generate digital content such as images, text, and the code. Recent progress in large-scale AI models has improved generative AI’s ability to understand intent and generate more realistic content. This text summarizes the history of generative models and components, recent advances in AI-generated content for text, images, and across modalities, as well as remaining challenges.

In recent years, Artificial Intelligence Generated Content (AIGC) has gained much attention beyond the computer science community, where the whole society is interested in the various content generation products built by large tech companies. Technically, AIGC refers to, given human instructions which could help teach and guide the model to complete the task, using Generative AI algorithms to form a content that satisfies the instruction. This generation process usually comprises two steps: extracting intent information from human instructions and generating content according to the extracted intentions.

Generative models have a long history of AI, dating to the 1950s. Early models like Hidden Markov Models and Gaussian Mixture Models generated simple data. Generative models saw major improvements in deep learning. In NLP, traditional sentence generation used N-gram language models, but these struggled with long sentences. Recurrent neural networks and Gated Recurrent Units enabled modeling longer dependencies, handling ~200 tokens. In CV, pre-deep learning image generation used hand-designed features with limited complexity and diversity. Generative Adversarial Networks and Variational Autoencoders enabled impressive image generation. Advances in generative models followed different paths but converged with transformers, introduced for NLP in 2017. Transformers dominate many generative models across domains. In NLP, large language models like BERT and GPT use transformers. In CV, Vision Transformers and Swin Transformers combine transformers and visual components for images.

The Big Data revolution has strained the capabilities of state-of-the-art electronic hardware, challenging engineers to rethink almost every aspect of the microchip. With ever more enormous data sets to store, search and analyze at increasing levels of complexity, these devices must become smaller, faster and more energy efficient to keep up with the pace of data innovation.

Ferroelectric field effect transistors (FE-FETs) are among the most intriguing answers to this challenge. Like traditional silicon-based transistors, FE-FETs are switches, turning on and off at incredible speed to communicate the 1s and 0s computers use to perform their operations.

But FE-FETs have an additional function that conventional transistors do not: their ferroelectric properties allow them to hold on to electrical charge.

With artificial intelligence poised to assist in profound scientific discoveries that will change the world, Cornell is leading a new $11.3 million center focused on human-AI collaboration that uses mathematics as a common language.

The Scientific Artificial Intelligence Center, or SciAI Center, is being launched with a grant from the Office of Naval Research and is led by Christopher J. Earls, professor of civil and environmental engineering at Cornell Engineering. Co-investigators include Nikolaos Bouklas, assistant professor of mechanical and aerospace engineering at Cornell Engineering; Anil Damle, assistant professor of computer science in the Cornell Ann S. Bowers College of Computing and Information Science; and Alex Townsend, associate professor of mathematics in the College of Arts and Sciences. All of the investigators are field faculty members of the Center for Applied Mathematics.

With the advance of AI systems – built with tangled webs of algorithms and trained on increasingly large sets of data – researchers fear AI’s inner workings will provide little insight into its uncanny ability to recognize patterns in data and make scientific predictions. Earls described it as a situation at odds with true scientific discovery.

A recent work introduces a cellular deconvolution method, MeDuSA, of estimating cell-state abundance along a one-dimensional trajectory from bulk RNA-seq data with fine resolution and high accuracy, enabling the characterization of cell-state transition in various biological processes.

Single-cell transcriptomic techniques continue to revolutionize the resolution of cell analysis, determining discrete cell types and cell states with continuous dynamic transitions that can be related to development and disease progression⁵. Cells in different states can be computationally ordered according to a pseudo-time series, or cell trajectory⁶. Both MeDuSA and another method, Cell Population Mapping (CPM)⁷, were developed to exploit the rich spectrum of single-cell reference profiles to estimate cell-state abundance in bulk RNA-seq data, which enables fine-resolution cellular deconvolution (Fig. 1b). Although CPM effectively tackles the issue of estimating the abundance of cells in different states, MeDuSA further improves the estimation accuracy by employing a LMM (see the equation in Fig. 1c) that takes into account both the cell state of interest (focal state) and the remaining cells of the same cell type (non-focal state) as well as the other cell types.

To show the capability of the OrganoidChip in enabling higher-resolution imaging, we used confocal microscopy for several organoids immobilized on the chip. Representative images show improved optical segmentation and the ability to resolve single cells within an organoid (Fig. 4 d). The co-localized EthD-1-and Hoechst-stained nuclei are resolvable and can potentially be used to increase the accuracy of viability measurements. Future implementation of 3D-segmentation using AI-assisted algorithms in the analysis pipeline can provide more accurate estimations of cellular viability in larger screens.

Next, we measured the effect of DOX treatment on the beating kinetics of cardiac organoids. To do this, we relied on calcium fluorescence imaging, as it has been shown to be a good approximation of the cardiomyocytes’ action potentials³². Calcium imaging proved beneficial for beating and contraction parameters since smaller beating portions cannot necessarily be detected from brightfield images, particularly when organoids have been compromised as a result of drug treatment.

When assessing drug effects, we observed some degree of variability in the spontaneous contractile behaviour and beating kinetics between cardiac organoids. Such variability often skews any averaged parameter value across organoids and does not reflect the effect of the treatment conditions on organoid health. To address this challenge, we tracked each individual organoid’s beating off-and on-chip. The drug-induced functionality results are therefore reported as averages of fractional changes of each individual organoid’s beating kinetics parameters, measured at 48 h post-treatment, on both the chamber slide and on the chip, relative to its pre-treatment value (Eq. 3).

An interdisciplinary team at the Advanced Quantum Testbed (AQT) at Lawrence Berkeley National Laboratory (Berkeley Lab) and the University of California, Berkeley’s Quantum Nanoelectronics Laboratory (QNL) achieved a technical breakthrough using qutrits—three-level systems—on a superconducting quantum processor.

The team successfully entangled two transmon qutrits with gate fidelities significantly higher than in previously reported works, thus getting closer to enabling ternary logic that can encode more information than their binary counterparts—qubits.

Published in Nature Communications in December 2022 and featured as an editor’s highlight, this experimental success pushes forward AQT’s qutrit research and development, including previous experimental successes published in 2021 in Physical Review X and Physical Review Letters. Ternary quantum information processors offer significant potential advantages in quantum simulation and error correction, as well as the ability to improve certain quantum algorithms and applications.

Apple’s Vision Pro headset will use a new type of dynamic random access memory, or DRAM, that has been custom designed to support Apple’s R1 input processing chip, reports The Korea Herald.

Apple Vision Pro is powered by a pair of chips. The main processor is the M2, which is responsible for processing content, running the visionOS operating system, executing computer vision algorithms, and providing graphical content.

Back in 1956, Denham Harman proposed that the aging is caused by the build up of oxidative damage to cells, and that this damage is caused by free radicals which have been produced during aerobic respiration [1]. Free radicals are u nstable atoms that have an unpaired electron, meaning a free radical is constantly on the look-out for an atom that has an electron it can pinch to fill the space. This makes them highly reactive, and when they steal atoms from your body’s cells, it is very damaging.

Longevity. Technology: As well as being generated in normal cell metabolism, free radicals can be acquired from external sources (pollution, cigarette smoke, radiation, medication, &c) and while the free radical theory of aging has been the subject of much debate [2], the understanding of the danger free radicals pose led to an increase in the public’s interest in superfoods, vitamins and minerals that were antioxidants – substances that have a spare electron which they are happy to give away to passing free radicals, thus removing them from the danger equation.

But before you reach for the blueberries, it is important to know that, as so often in biology, the story is not black and white. Like a misunderstood cartoon villain, free radicals have a beneficial side, too – albeit in moderation. Free radicals generated by the cell’s mitochondria are beneficial in wound-healing, and others elsewhere act as important signal substances. Used as weapons by the body’s defense system, free radicals destroy invading pathogenic microbes to prevent disease.