Nous Group

Looking forward to a great day with Clutch Events at the Melbourne Data Protection & Security Summit on Thursday. Should be a great day addressing some of the hot topics facing all organisations today.

New blog post -- Introducing: fable.gam , a new forecasting option Those who know me or read my little blog know that I am: (i) a massive fan of generalised additive models (GAMs); and (ii) a massive fan of Rob Hyndman, Mitchell O'Hara-Wild , and Earo Wang's `tidyverts` ecosystem for doing tidy time-series analysis, particularly using the `fable` R package for forecasting. `fable` (and the underlying `fabletools`) provide an excellent forecasting workflow, including handling transformations, fitting models, computing accuracy metrics, combining models, and a whole host of other useful stuff. Recently, I have been experimenting with trying to combine the two by developing an extension package in R to fable called ` fable.gam `. While there are certainly a lot of strong considerations to make when using GAMs for time-series analysis (such as the implications for forecasting out a smooth term fit for the trend), they do provide a lot of scope for flexible modelling of both nonlinear trends and seasonality through cyclic cubic splines (shout out to Gavin Simpson whose 'modelling seasonal data with GAMs' blog post from years ago really stuck with me), as well as handling non-Gaussian variables. There is a lot of work I still have to do around the user interface for specifying smooth functions and model diagnostics etc., but the results in the blog post suggest that maybe there is something to this sort of novel way of doing short-range forecasting (I probably wouldn't want to use smooth functions for longer time horizons). I show that leveraging R's incredible `mgcv` package for fitting GAMs under-the-hood in ` fable.gam ` can outperform the closest open-source model to this idea which is Facebook's `Prophet`. However, I do need to think about Gavin Simpson's 'Extrapolating with B splines' blog post some more in relation to the smoothing basis -- especially given the discrepancies between the thin plate regression spline and Gaussian process I observed in my analysis. Please give the post a read at the link below if you are interested! https://lnkd.in/gZCJvs_u #statistics #timeseries #rlang #rstats #forecasting #datascience

I've been talking about getting your leaders more visible lately. But it's not for the fun of it. For organisations, it's part of a broader channel strategy. Your corporate channel reaches one audience. Your website traffic is probably down. The algorithm moved on. But your people? Your people reach different networks. Different contexts where they're actually trusted. Five leaders sharing from their own channels isn't five times your corporate reach, it's exponentially more. Because each one has a different network. Different credibility. Different audience that pays attention to them specifically. This is a force multiplier for your organization's thinking! Look at the gap between these two models. That's where I think organisations can really harness the smart people to get their best ideas in front of the right people.

"It's a tool to take care of low-level thinking, not a substitute for your skills and expertise in public service delivery. " This is pretty reductive conclusion by The Mandarin emailed to their large audience of Australians working in and around the public sector. Cognitive offloading or atrophy from AI is a real risk, but the risk management approach is not to limit AI to low-level thinking. The value of AI for knowledge work depends on whether it makes us think more or less. The tools are agnostic. We can build AI that provokes or AI that sedates. This is a choice that we can make. Reductive messages like these hold us all back from doing the bigger thinking that is required to design AI tools and services that produce better, not just faster, outcomes for citizens. Am I off track here? Paul Hubbard Christina Wiremu-Brook Michael Rathjen The Hon. Victor Dominello ?

Energy governance sits at the centre of Australia’s transition, and it’s becoming increasingly complex. Next week, I’ll be joining a panel for The Energy to discuss how Australia’s energy system is governed, where accountability should sit, and what needs to change to better support whole‑of‑system outcomes. The webinar will cover: - Key findings from the Power Dynamics white paper - The roles of market institutions and where governing authority should rest - How to better incorporate demand‑side expertise - The importance of data, research and evidence in policy design I’m pleased to be part of a panel alongside Dani Alexander (UNSW Energy Institute), Drew Clarke (former Commonwealth energy secretary and Chair of AEMO), Rob Murray-Leach (Energy policy expert) and Alison Reeve (Grattan Institute). This session will be relevant for policymakers, market bodies, industry leaders, investors and network service providers. Register free here: https://lnkd.in/gTPac3Cj