Last week I had my first experience of the STOR-i conference. This is a two day event at which selection of academics, industry professionals and STOR-i PhD students show off their work. The line-up this year was great, with a broad range of interesting and thought provoking presentations. It would be infeasible to discuss all of them, so I have picked out a couple that really resonated with me.
In a few slides of the presentation the speaker explained how EDF go about modelling the relationship between temperature and energy consumption. This I found very interesting, since in my studies up till now I have worked mostly with linear models, and haven't seen many examples of non-linear dependencies and how they can be modeled. The relationship Audrey was discussing is, however, a very intuitive example of non-linear dependence. When the temperature is very low many households will be using heating to keep warm, likewise, when it is very hot, air conditioning will be getting used to keep things cooler. Between theses two extremes the consumption is much lower, resulting in what will quite clearly be a non-linear relationship.
With many years experience in eCommerce, including one year at PayPal and four with eBay, she shared some of her advice on communicating with users effectively. Although it was borne from time in a different area of expertise than mine, this advice seems quite applicable to the work I and others do here at STOR-i.
I feel this perspective is applicable to our work here at STOR-i. The majority of research we do is in partnership with industry, with practical applications and problem solving being the ultimate aim. You can view anyone interested in our work as the 'users', this could be employees of these industrial partners or fellow academics, for example. Care must then be taken in the communication to our users about clever work we have been doing. If this is done badly, then no matter the strength of the work done, it will have less impact.
Forecasting energy consumption in France
This talk was given by Audrey Lagache from EDF. The topic of discussion was their use of forecasting techniques to aid their operations. Since electricity is not easily stored, it is necessary for EDF to predict consumer usage in order to match energy provision to demand, avoiding any unwanted costs.In a few slides of the presentation the speaker explained how EDF go about modelling the relationship between temperature and energy consumption. This I found very interesting, since in my studies up till now I have worked mostly with linear models, and haven't seen many examples of non-linear dependencies and how they can be modeled. The relationship Audrey was discussing is, however, a very intuitive example of non-linear dependence. When the temperature is very low many households will be using heating to keep warm, likewise, when it is very hot, air conditioning will be getting used to keep things cooler. Between theses two extremes the consumption is much lower, resulting in what will quite clearly be a non-linear relationship.
The question is then how to model this relationship. At EDF they use Generalized Additive Models (GAM), a form of Generlized Linear Model in which one looks to approximate the linear predictor with a sum of smooth functions. The set up looks something like:
\( g(\mu) = b_0 + f_1(x_1) + f_2(x_2) + \dots + f_p(x_p)\)
\(\mathrm{where} \quad \mu = \mathrm{E} \left[ \mathbf{y} \right].\)
This talk was given by Elisabeth Ling of Elsevier. According to their website, they are "a global information analytics business that helps institutions and professionals advance healthcare, open science and improve performance for the benefit of humanity. " The talk centered around their use of data science for improving the tools they offer to help researchers with their workflows.
\(\mathrm{where} \quad \mu = \mathrm{E} \left[ \mathbf{y} \right].\)
The \(x_i\) here are the covariates , \(\mathbf{y}\) is the response variable, the \(f_i\) are the mentioned smooth functions and \(g\) is a link function. From what I understand, since there are a wide range of smooth functions to choose from, there is more flexibility in the relationships that can be captured. Hence, the non-linear dependencies seen can be modeled.
Improving researcher tools with data science
With many years experience in eCommerce, including one year at PayPal and four with eBay, she shared some of her advice on communicating with users effectively. Although it was borne from time in a different area of expertise than mine, this advice seems quite applicable to the work I and others do here at STOR-i.
The first thing discussed was the need for good quality front-end website design that does not inhibit any data science work going on in the back-ground. For example, Elsevier would like to provide smart recommendations to users, helping them sift through the large amounts of accessible information to find things which will be most useful to them. Suppose they have done some very clever data science work that can produce a top 5 list of recommendations for the user, which will get displayed on the current page. If the placement of these recommendations on the web page is poor, or if the layout makes it hard to read, then users are unlikely to pay any attention to this feature, rendering all behind the scenes work useless. Therefore, just as much time must be spent thinking about communicating with users as on the underlying technicalities. This was summed up in a quote: "Users are like gold dust, take care with how you communicate with them.""Users are like gold dust, take care in how you communicate with them."
I feel this perspective is applicable to our work here at STOR-i. The majority of research we do is in partnership with industry, with practical applications and problem solving being the ultimate aim. You can view anyone interested in our work as the 'users', this could be employees of these industrial partners or fellow academics, for example. Care must then be taken in the communication to our users about clever work we have been doing. If this is done badly, then no matter the strength of the work done, it will have less impact.
"Fall in love with the problem, not the solution."In one of the final slides, the speaker presented the above quote. It centers around the idea that work should be done with future application and impact in mind. For example, if you focus too much time on producing a really neat solution to a problem, even though this solution may not be that applicable, then it may be a waste of time. Perhaps one should instead maintain the focus on the problem itself and find solutions to fit. This is likely to be easier said than done, especially when solutions aren't jumping out at you. However, I feel it is a interesting view on things.
Comments
Post a Comment