ML - The way the world works - analyzing how things work

Quantum computing

Nlp generative computer code

Weekly thoughts

Recap of thoughts

veritone wants to change the world, to build something never built before. veritone wants to make an impact. veritone analyzed video and detected criminal behavior or criminals and help police capture them. veritone built a platform that processes three to four million hours of content every quarter. veritone team members ideas are welcome. may the best idea win. veritone help catalog giants video content making it searchable and accessible by the fans. veritone offers text to speech and it can analyze voice for pattern variance. veritone ai to analyze energy grid performance. veritone has built an ai operating system. the path to ai is collaberation. human in the loop provides the cognitive processing. veritone can learn customer voices. the call is transcribed in text and then ai analyzes the text. veritones operating system platform ai domain is : vision, speech, data, biometrics, audio, and text ingestion and analysis.

aibrain strives for human like intelligence. Cognitive reason for humans makes us want to learn about the world. People remember states or memories about the world. increased computing power of mobile phones have allow consumers to enjoy ai like siri. ai has been steadily evolving rather than revolutionary. ai has an enormous difficulty to understand human language. aibrain wants to deliver ai intelligent characters. ai will take care of mundane tasks. human beings will work on imagative work. cars and airplanes are tools used by humans likewise ai will be tool to augment human intelligence. https://lnkd.in/g2-mZrh

cerebras creates a chip with over a trillion transistors. cerebras can run deep learning algorithms faster than other infrastructure. cerebras has 400,000 cores in a single chip.  The main chip is called cs-1. cs-1 has redundant cores for failover controlled by software. cs-1 has 18 gigabytes of memory and 9 pentabytes of memory bandwidth. Each processor has its own memory. cs1 has high speed communications between the cores.

deepset works to get more meaningful search results. Deepset uses transfer learning, language models, and question and answer to drive search results. Making sense of text data. deepset is an open source company. It uses natural language processing to answer questions using bert. MiniLM is a small version of BERT. deepset uses extractive questioning and answering where the answer must be part of the text by one span. This is an energy problem. There is a large set of potential predictions. SQuAD has become the default dataset. BERT is a feedforward network predicting the start and end token of the extractive answer then maps to the word dictionary to produce and answer. Industry wants question and answer for its data. There is a growing number of knowledge workers spending hours and hours reading text. Knowledge workers are use to web search and enterprise search. Scaling involves gathering a large number of documents. The road map include generative question and answering and sythentic reasoning.

soul machine is trying to humanize ai called sam. sam simulates an ai brain reacting to noise, talking, and information. sam has facial expressions. sam can interact back with the user. Sam can remember the conversation. Sam uses the web cam too see the user. The facial gestures makes sam seem more human. Ai will help people interact better with things.   https://lnkd.in/gfYQiJf

Emteq - facial emotion recognition www.emteq.net. Emteq software and hardware measure muscle tensions in the face then send those signals to a deep learning network that receives the inputs and outputs facial expressions in an avatar.  eye tracking and blinking is also tracked. https://lnkd.in/g7X7Kht