Artificial Intelligence Suisse

22/07/2021

An that interprets the brainwaves of a man who can't speak and turns them into conversations - Researchers at UC San Francisco and UC Berkeley trained a system to interpret electrical impulses from the brain of a man who had lost the ability to speak 15 years ago and displayed them as words on a video screen.

A stack of three LSTMs detected portions of brain activity related to speech. An ensemble of 10 convolutional gated recurrent unit models classified speech signals as one of the 50 words. An n-gram language model predicted the probability that a given word would come next. A custom Viterbi decoder, an algorithm often used in communications that are subject to transmission errors, determined the most likely of the 50 phrases based on the models’ output.

Details/research: https://www.nejm.org/doi/full/10.1056/NEJMoa2027540

On LSTM networks (pdf paper):http://www.bioinf.jku.at/publications/older/2604.pdf

Viterbi decoder: https://en.wikipedia.org/wiki/Viterbi_decoder

15/07/2021

For years, natural language generation (NLG) researchers have been recruiting humans to evaluate their models’ text outputs. But in a new study, University of Washington and Allen Institute for Artificial Intelligence researchers argue that untrained humans are not the natural language experts we’d like to think they are.

For years, natural language generation (NLG) researchers have been recruiting humans to evaluate their models’ text outputs. This practice…

14/07/2021

Next frontier for GPT-3: Automating coding of complex problems

OpenAI’s GPT-3 has made headlines since its release last June. Recent integrations have revealed this large language model’s additional…

12/07/2021

Tips on how to win your next -hackathon. Are you ready?

My story of entering hackathons and how I learned to win at a high rate

12/07/2021

Cornell and Harvard University Researchers Develops Correlation Convolutional Neural Networks (CCNN): To Determine Which Correlations Are Most Important

Details: https://www.marktechpost.com/2021/07/10/cornell-and-harvard-university-researchers-develops-correlation-convolutional-neural-networks-ccnn-to-determine-which-correlations-are-most-important/
Paper:https://www.nature.com/articles/s41467-021-23952-w.pdf

11/07/2021

Interested in the AGI capability roadmap? - This maturity model continues to be an excellent map to guide your understanding of the progress towards artificial general intelligence.

Continue here: https://medium.com/intuitionmachine/revisiting-the-agi-capability-roadmap-a89a533e5e4c

11/07/2021

The expressive power of GNNs mainly reflects their capability to distinguish whether two given graphs are isomorphic or not. Recent studies however have shown that GNNs struggle with long-range interactions and lack a principled way to model higher-order structures, limiting their expressive power.

A new paper from the University of Cambridge, Imperial College London & Twitter, UCLA, MPI-MIS, and SJTU & UNSW proposes CW Networks, a message-passing method that delivers better expressivity.

The paper Weisfeiler and Lehman Go Cellular: CW Networks: https://arxiv.org/abs/2106.12575

05/07/2021

The jobs ahead --- If GPT-3 can write code, could it end coding? If it can write poetry, fiction, or songs, could it end writing? It seems we’d have to rethink completely our perspective on the future that’s falling onto us.

OpenAI’s CEO made a prediction that could change everything.

05/07/2021

A marketer's dream is advertising fully customized and independent from a user's location on live TV. Envision a European soccer competition where people see different ads just based on their individual needs.

05/07/2021

The BrickIt App scans LEGO bricks to suggest things you can build and I think it's an amazing example of what's possible when you combine with