Alexa Self Learning to Correct Mistakes

In all conversation there is adjustments of our interactions.Good piece here that shows how this is being proposed for a common assistant.

Amazon Uses Self-Learning to Teach Alexa to Correct its Own Mistakes

The digital assistant incorporates a reformulation engine that can learn to correct responses in real time based on customer interactions .    By Jesus Rodriguez in Towards Data Science

 Digital assistant such as Alexa, Siri, Cortana or the Google Assistant are some of the best examples of mainstream adoption of artificial intelligence(AI) technologies. These assistants are getting more prevalent and tackling new domain-specific tasks which makes the maintenance of their underlying AI particularly challenging. The traditional approach to build digital assistant has been based on natural language understanding(NLU) and automatic speech recognition(ASR) methods which relied on annotated datasets. Recently, the Amazon Alexa team published a paper proposing a self-learning method to allow Alexa correct mistakes while interacting with users.

The rapid evolution of language and speech AI methods have made the promise of digital assistants a reality. These AI methods have become a common component of any deep learning framework allowing any developer to build fairly sophisticated conversational agents. However, the challenges are very different when operating at the scale of a digital assistant like Alexa. Typically, the accuracy of the machine learning models in these conversational agents is improved by manually transcribing ... '

Unifying Logical and Statistical AI With Markov Logic

As AI practitioners in the enterprise we understood this early on.   You need to know the results of statistical analysis AND the ability to link them usefully to logical decision making.   Sometimes easy,  sometimes not   Thus approaches like decision trees based on statistical data became popular for our team.   We understood too that Markov methods could provide the framework for providing this, so we experimented with them.  In both cases the results were also relatively transparent.    This unification can also outline way that humans will interact with the AI.  Research on the idea was going on then and is still now.  Below gives you a good update.  Starts basic and gets technical.

Unifying Logical and Statistical AI with Markov Logic
By Pedro Domingos, Daniel Lowd 
Communications of the ACM, July 2019, Vol. 62 No. 7, Pages 74-83    10.1145/3241978

For many years, the two dominant paradigms in artificial intelligence (AI) have been logical AI and statistical AI. Logical AI uses first-order logic and related representations to capture complex relationships and knowledge about the world. However, logic-based approaches are often too brittle to handle the uncertainty and noise present in many applications. Statistical AI uses probabilistic representations such as probabilistic graphical models to capture uncertainty. However, graphical models only represent distributions over propositional universes and must be customized to handle relational domains. As a result, expressing complex concepts and relationships in graphical models is often difficult and labor-intensive.  .... "   (  Full Technical paper)

Video intro to the concept (technical): 





Alchemy Language, mentioned in the above talk:

https://alchemy.cs.washington.edu/
Alchemy: Open Source AI
Welcome to the Alchemy system! Alchemy is a software package providing a series of algorithms for statistical relational learning and probabilistic logic inference, based on the Markov logic representation. Alchemy allows you to easily develop a wide range of AI applications, including: .... " 

Bots and Spam

Very interesting historical and technical view of Spam.   First the history and then the look at how the technical model of a markov chain is involved. Great example to think about how state models work.  And how these models start to look like the basic intelligence of Bots.

The claim is made that the first commercial Spam was sent in 1994.  That made me think, I am sure I saw Spam on groups well before then.  But then it may hang on what 'commercial' means, and the volume involved. Well worth a read from the technical perspective, has given me a few ideas already.

Dynamic Programming in Health Care

The OBAIS department at the Lindner College of Business, University of Cincinnati, invites you to attend the following research seminar.

Date & Time: Friday, April 10, 2015, 1:30 PM
Location : UC Carl H. Lindner Hall, Room 219
Speaker: Dr. Steven Shechter, Associate Professor, Sauder School of Business, University of British Columbia

TOPIC: Approximate Dynamic Programming in Health Care: Linear Programming and Simulation Based Approaches

Abstract: In this talk, I will discuss two distinct health care projects that are naturally modeled  as Markov decision processes, but which face the curse of dimensionality.  The first problem concerns blood issuing policies for hospital blood banks.  We describe our use of LP-­‐based ADP approaches for finding policies that outperform existing policies regarding the inherent freshness/shortage trade-­‐offs.  The second problem concerns the effective allocation of operating room time for elective surgeries at the British Columbia Children’s Hospital.  Here, we apply simulation-­‐based methods for dealing with the complex state dynamics as well as the large  state space of the underlying MDP.  

Via Uday S. Rao, Associate Professor
Operations, Business Analytics, and Information Systems (OBAIS)
Email: raous@ucmail.uc.edu     @UCBusAnalytics

Wikibrains Semantic Visualization Tool

Former colleague writes  Leon Markovitz writes:
" ... Currently, I'm in TLV helping develop wikibrains.com, a cool tool to quickly create data visuals. We try to keep the UI as simple and minimalist as possible. ... Every connection made is saved, to then recommend it to the next users mapping similar topics. ... It's creating a semantic web, one map at a time. ... "

Hidden Markov and More

A non technical description of a very technical set of methods for forecasting. Baum-Welch and Hidden Markov.  Don't know much about this direction but looks to be interesting.
New ideas are sprouting up. Not too much here on the how to use this method, but looking further.

Wikipedia to Make Computers Smarter?

Wikipedia Used to Maker Computers Smarter
Fascinating note of work at the Technion. The idea of background or common-sense knowledge is an inportant one in artificial intelligence. Since a Wikis knowledge is potentially ever-changing, and we know the WP has led to much debate over the correctness of its knowledge, how would such an approach work?

" ... The program created by the research team uses a concept database, constructed from Wikipedia, to understand single words and phrases. The type of "background knowledge" that the researchers want computers to utilize is a vital part of human problem solving ability, "but we [previously] didn't know how to have computers access such knowledge," says Technion Faculty of Computer Science researcher Shaul Markovitch. Whereas current programs simply treat documents as a group of words, the new system aims to understand the meaning of the words it encounters..."