Tech News Review Week 9-14 September 2020


An Army of Microscopic Robots Is Ready to Patrol Your Body

The actuators of these microrobots can be operated by the low-power electric currents that typically flow through electronic circuits, allowing the most advanced micro-electronics research to be incorporated into modules so small they can’t be seen by the human eye. Along with gene therapies, microrobots are the future of medicine and could be the next trillion-dollar industry within the next 20 years.


Ab8 COVID-19 Drug Breakthrough: Tiny Antibody Component Completely Neutralizes the SARS-CoV-2 Virus

Monoclonal antibodies are a far more superior strategy against COVID-19 than vaccines or antiviral drugs, they are however very expensive and slower to manufacture, which is possible in plants costing billions of euros, making this solution harder to scale.


Gene-Hacked “Bodybuilder” Mice Stayed Ripped on Space Station

The envisaged use cases for astronauts may be far-stretched and not so useful, however, the future medical applications of this gene therapy could mean that older adults or individuals who are bedridden or wheelchair-bound from illness will find a cure.



This Microchip Has Its Own Built-In Cooling System

Single thread CPU performance has plateaued during the last 10 years, exhibiting only 3-5% incremental increases in performance every year, however, Built-In Cooling System could accelerate progress, and furthermore allow for more dense and powerful electronics into constrained platforms like drones or wearables. Especially autonomous cars will thus be capable of having a built-in supercomputer, even hundreds of time more powerful than current onboard systems.


Quantum computing

Extracting order from a quantum measurement finally shown experimentally

The outcomes of this fundamental quantum mechanics research allow to extract order from the largely disordered system, providing a general tool to engineer the state of the system, essential for future quantum technologies, like quantum computers.


Machine Learning

Descending through a Crowded Valley —Benchmarking Deep Learning Optimizers

Further improvement in the optimizer state of the art used by machine learning models is essential for the evolution of AI. Any progress in this field makes AI more powerful and efficient in terms of energy and hardware costs.


[2006.10029] Big Self-Supervised Models are Strong Semi-Supervised Learners

Large models pre-trained in an unsupervised fashion and then fine-tuned on a smaller corpus of labeled data have achieved spectacular results in natural language processing. New research pushes forward with a similar approach to computer vision.


DeepSpeed: Extreme-scale model training for everyone – Microsoft Research

DeepSpeed adds four new system technologies making model training with billions to trillions of parameters feasible with contemporary hardware. The underlying models are compatible with extremely long input sequences and can run on hardware systems with a single GPU, high-end clusters with thousands of GPUs, or low-end clusters with very slow ethernet networks. This is is a very positive development for the democratization of advanced AI research.


[2009.04765] Brain2Word: Decoding Brain Activity for Language Generation

The underlying machine learning model maps an fMRI scan to the corresponding word within a fixed vocabulary. The results from this research could one day lead to an AI system that translates brain activities into coherent text.




New Military Drone Fits in Backpack, Can Carry Lasers, Radio Jammers, Weapons

The future of military warfare will be 100% unmanned, and it is already here, as shown by all the demonstrations of novel robotic and drone systems that emerge every week.