Interview with Liang Downey, IBM

liang downey

Liang Downey works for IBM as a Business Development Executive for IBM’s New Energy and Environment Group. She promotes emerging solutions that leverage IOT, analytics and cognitive computing to better balance the distributed energy supply and demand, eliminate energy waste and reduce GHG emission. Her professional experience spans over 20 years in consulting, program management and emerging technology development across multiple industries. For 5 years at a renewable energy start up, she helped the company grow, from business strategy and partnership to licensing company’s IP.

Liang joined a group of volunteers to support the IEEE Humanitarian Technology Challenge (HTC) in 2009. HTC was designed to enable remote villages in regions of the world that lack access to electricity by deploying mobile solar trailers built-in with stationary battery, home batteries, as well as LED lightings and phone charger, to light up and digitize dark homes. HTC has since evolved into IEEE Smart Village initiative today.

Ms. Downey earned a Master Degree in Electrical Engineering from Clarkson University in New York, with GPA 4.0/4.0. She received her MBA in Finance from Wayne State University in December 2014.

In this interview, Downey answers questions regarding her IEEE Smart Grid webinar, “Unlocking the Value of IoT: A Cognitive Energy System Future.” To view this webinar on-demand, click here.

 

QUESTION: What do you see as activities research on smart grid within universities?

ANSWER: Battery research, Grid situation awareness simulation/monitoring software, Demand response programs and Cognitive controls on distributed energy resources.

QUESTION: Where can we see cognitive learning by the next decade? Will it play a major role in our day to day life?

ANSWER: Learning will be personalized based on interaction between educator and learner aided with machine learning. A virtual educator, augmenting traditional teaching, will be able to interact with learners based on direct feedback to improve knowledge transfer.

QUESTION: Most of utility companies applaud the models of the modern grid/utility industry while they're struggling with the "unfair" competition from the renewable energy as well as the push toward the Distributed Energy Resources, how many of them really want it?

ANSWER: Distributed Energy Resources (DER) offer flexibility, location advantage as well as sustainability benefits. It is a valid alternative to help utilities defer costly grid upgrade, it also serves as a catalyst to spur economic activities by the peer to peer energy trading. On-site DER and the aggregation of DERs (Virtual Power Plant) gives a city, utility and provider of energy the flexibility to respond to population and energy needs changes.

Utilities are the indispensable provider of "the grid and the future grid," and that return-on-capital model will still apply to them. Batteries, volt/var optimization, state estimation, and grid planning still fall under that model. The "unfairness" comes from trying to justify upping the rate case to handle the grid impacts caused by DER, but on the whole I don't see any grave concerns on their part.

QUESTION: Could you please mention some resources/tools to start/learn Academic-research on smart grids

ANSWER: Universities have used Cplex to solve energy optimization problems.

Click here to access the IBM Research activities in Smart Grid/Energy Efficiency.

QUESTION: How are cognitive solutions related to analytic solutions?

ANSWER: Machine Learning takes advantage of today's modern and very high speed computational processing ability to solve Deep Learning algorithms, such as Convolutional Neural Networks (as used in Visual Recognition, for instance). If you consider machine learning the base level of analytics, artificial intelligence is the next level. At the top (the 3rd level) it is the Cognitive analytics, mimic the human-like abilities to hear, to see and to understand natural languages and to learn, so to be able to grasp concept, form hypothesis and made optimized decision.

To understand natural languages, the “natural language systems” need to be able to parse and interpret spoken language; applications include direct machine-to-human interactions.

Cognitive system is a more complex type of analytics.

QUESTION: Which machine learning methods are suitable for handling smart grid data?

ANSWER: Supervised learning is more suitable

QUESTION: How does the university conducting research involve so much data which is mainly owned by the utility?

ANSWER: University can form partnership with utilities if your research can benefit utilities. Utilities are generally open to sharing data for research purposes, but expect to sign a non-disclosure agreement that will limit or prohibit sharing of data outside the entities involved in the project. Expect too that the utility will want to own some/all of any intellectual property that is generated as part of the research.

QUESTION: Can you explain again how a cognitive system is different from neural network?

See answer to the question above. Neural network is a type of AI system, it is the mechanics to support the development of a cognitive system.

QUESTION: What is Watson, is it accessible to university? and what is the difference between Watson and other platform such as matlab, azure and so on.

ANSWER: Watson is the brand to represent the IBM platform that enable users to build cognitive applications. It has several layers of building blocks. On the bottom it is the DATA layer delivered by the “Watson Data Platform” to allow organizations to integrate/curate and ingest difference types of data, structured data, unstructured data, internal or external data, real-time or off-line data. Then you can apply AI to the data. Examples of AI include Conversational AI, Visual Recognition AI, Discovery AI, Speech AI, Natural Language understanding AI, Knowledge Query, Tone Analysis, and so on. These AIs are used in combination to build business applications that are hosted in the IBM Cloud.

Yes, Watson is accessible by universities. Click here for a link to the IBM Global University Program with links to research and student assets.

Matlab is mathematical simulation/analysis tool for engineering systems, where the system behavior is governed by natural laws. When the relationship between system input and output cannot be defined by equations or algorithm, machine learning can come into play to fill the gap by learning the input/output behavior. This is where Watson machine learning differs from tools such as Matlab.

Azure is Microsoft’s cloud offering. Watson, Cognitive solution and Watson AI can be hosted in IBM’s Cloud center.

QUESTION: Which concept is predominant in IoT machine learning, SVM or ANN?

ANSWER: Support Vector Machine (SVM) or Artificial Neural Network (ANN)? It depends, both can be considered classification methods; selection is a matter for data science methods.

QUESTION: Does Watson have a use case in the manufacturing industry besides utility?

ANSWER: Yes, there are numerous Watson use cases in manufacturing industry to help with plant optimization, decision support, and quality controls. Example include Watson Plant Advisor, Watson Equipment Advisor. Preventative maintenance and early warning systems.

QUESTION: How will a system be regulated and/or surveyed/protected so that it does not have inputs to "hate" or "target"?

ANSWER: It should be part of the “Governance” model to be embedded in the process of such system development. Any given opportunities provided by any innovation since history always come with risks. Internet connects people, organization and communities, government and citizens, but hackers and evil doers take advantage of such network to for exploitation. AI system is of no exception. Bad possibilities will not stop innovation but risks must be addressed. Here is an article on “Partnership on AI” formed by Google, Facebook, Amazon, IBM and Microsoft, which aims to set societal and ethical best practice for artificial intelligence research:
https://www.theguardian.com/technology/2016/sep/28/google-facebook-amazon-ibm-microsoft-partnership-on-ai-tech-firms.

QUESTION: At the end, the use cases are really interesting, are there any references for these applications?

Here is an example:
e2-Diagnoser: A System for Monitoring, Forecasting and Diagnosing Energy Usage Cognitive Building Operations

QUESTION: How do you consider system expansion into your cognitive learning?

ANSWER: That is an advantage of a Learning system as its accuracy and efficiency increase over time, it matures with every user/system interaction and output. It never stops learning thus a continues system of expansion.

QUESTION: Data based approaches often fail when there is any uncertain operation scenario. Say for fault location, different fault may occur. Trained model will be for some faults. What if actual fault is not anywhere close to trained cases?

ANSWER: To address these types of concerns requires a "triad" of project participants: a business owner (the person who wants to use cognitive to solve some business problem), the data scientist (the expert to invoke predictive analytics or cognitive capabilities), and a deep subject matter expert, one who can anticipate complex scenarios and unintended consequences, and that can drive requirements into the effort. New data sets will be fed into the model to improve the model accuracy.

QUESTION: Will the machine language "R" help in IoT environment of smart grid?-

ANSWER: I believe so. R is a package with broad, open source analytics capabilities. It is not the only path however; Python offers access to many analytic capabilities; there are also commercial entities (such as IBM Watson, Google, Microsoft) for cognitive, and open source too (Caffe, DeepLearning4j) and so forth. It is a big universe.