Recent

ads

Header Ads

Radar trends to watch: November 2020

Perhaps the most important event this month isn’t technical, but the start of the US Justice Dept.’s lawsuit against Google. That will certainly play out over years rather than months, but it’s significance is less about this particular case than the idea that legal and regulatory systems will play a large role in the evolution of technology in the US. In the short term, it’s worth watching the CPPA, GDPR, California’s Props 22 and 24, and FCC interference with social media’s enforcement of rules around community behavior.  Long term, this is only the beginning.

Artificial Intelligence and Machine Learning

  • Partial differential equations are the key to a number of difficult and important problems. In a surprising breakthrough, it’s been shown that deep learning can be used to solve PDEs, and that they are orders of magnitude faster than typical numerical methods.
  • Agence is a dynamic film/multiplayer VR game with intelligent agents. AI might not be what pushes VR to commercial success, but it will certainly be a part.
  • Towards more trustworthy models: Research on models that can take into account physical laws, error, and missing information makes AI more trustworthy.
  • Less than one shot learning”: when the number of classes is larger than the number of sample. The key idea here is “soft labelling,” which allows labels to represent characteristics shared between multiple items in the sample.
  • Very small neural networks adapted from worms’ nervous systems are able to perform tasks as well as networks thousands of times larger.
  • A device that can mimic the behavior of a neuron might eventually make it possible to emulate a brain, without requiring a lot of power.
  • AI is not magic, even when it is effective. Integrating AI into medical practice requires work–serious human work in rethinking social structures, communications, and hierarchies. This work frequently falls to the nursing staff, whose contribution is often undervalued.
  • Awful-AI is a GitHub repo for tracking “awful uses of AI.”  Autonomous weapons, racist AI, social credit, and scams: they’re all here. It would be funny if it wasn’t sad (and real).
  • Pattern-exploited training (PET) is a new NLP technique that, on some benchmarks, exceeds GPT-3 performance while only using 223 million parameters (as opposed to 175 billion). While 223 million is still a large number, that’s a factor of almost 1000 less than GPT-3.
  • Does explainable AI actually increase trust?  Perhaps not. Explanations can be distortions (lies).  Auditing is a more trustworthy approach.

Infrastructure and Operations

  • Microsoft’s Open Source Dapr Distributed Runtime is an abstraction layer on top of Kubernetes (and much more) to simplify building software that runs in a distributed multi-cloud environment.
  • Cloud waste: How much cloud spending is actually used? Perhaps as much as 45% is wasted on overprovisioning, unexpected costs, and resources that go unused.
  • Building continuously resilient systems starts with chaos engineering: Adrian Cockroft on failover theater, the cloud, and reliability.

Programming

  • An all no-code startup: Yes, it’s possible.  It’s time to start thinking about the no-code “stack.”
  • The kernel within the Linux kernel: eBPF can run sandboxed programs within the kernel’s memory space; it’s a convenient way to write kernel extensions and, conceivably, to replace much of the existing kernel with an eBPF-based microkernel.
  • It’s worth reading @sogrady’s thoughts on how developer experience needs to change in order to build more tightly integrated software across multiple platforms.

Security and Privacy

Economies

  • Indoor farming, food security, climate and COVID in Singapore: how do you guarantee a food supply in the face of supply chain breakdowns and an increasingly erratic climate, in a densely populated country with almost no arable land?
  • Towards a digital Euro: The European Central Bank is taking the first steps towards a digital version of the Euro.  It’s behind China (but ahead of the US) in its steps towards digital currency.

Hardware

  • Quantum fast fourier transform (QFFT): Want to reinvent digital signal processing on quantum computers? This is another game changer, and one of the few classical algorithms that have been reinvented for quantum computers.
  • Multi-state memory: not just zeros and ones. Moore’s law applies to memory, too, and the best way to get around the fundamental limitation (ever-smaller features) might be to design memory that goes beyond binary. Multi-state memory might also be a big step forward in neuromorphic computing.
  • Printing sensors directly on the body without heat: Who needs an iPhone? These sensors could be revolutionary for tasks like COVID monitoring and personalized medicine.
  • Robotic fabric: Developers have made fabric that can change its shape programmably.  Robotic clothes?

Web

  • Do Not Track failed. Global Privacy Control (GPC) is essentially Do Not Track with legal teeth. Will it succeed?  Currently implemented by the Brave browser and some plugins for other browsers.
  • Firefox’s campaign to “Unfck the Internet” gives us a reason to return to Firefox.  The campaign is built around a set of plugins for controlling political ads, social media surveillance, warning others about inappropriate YouTube recommendations, and more. Putting social media control in the browser? It might work.


from Radar https://ift.tt/35VvzmG
Radar trends to watch: November 2020 Radar trends to watch: November 2020 Reviewed by US Tech News on November 02, 2020 Rating: 5

No comments:

Follow Us

ads
Powered by Blogger.