Lunch Colloquium
Broadly accessible Overview talk

12:00-12:50 : S. Keshav, Cambridge — Is There Hope for the Climate?

With what seems to be a perpetually growing dependency on oil despite crisis after crisis, a global retrenchment from clean energy commitments, and a steadily rising carbon dioxide concentration in the atmosphere, climate activists and the general public alike seem to have little reason for hope. Are we consigned to a fate of climate extremes and human suffering? In this lecture, I will argue that rapid advances in renewable energy sources such as solar and wind, combined with the declining cost of storage, do give us reason for hope. Critically, computer science and computer scientists will play a huge role in the future energy system, making them essential to its infrastructure. I will outline this role and describe some open research challenges that await resolution.

Workshop

13:00-13:45 : Jessica Jewell, Chalmers — Probabilistic projections of global wind and solar power growth

Despite the recent surge of wind and solar power, both technologies need to accelerate to meet climate goals. Yet, there are no robust methods to assess the likelihood of such acceleration. Here we show that renewable energy deployment follows a recurring pattern across countries with prolonged periods of relatively steady growth punctuated by growth pulses. Based on this insight and on observed growth trajectories in early adopting countries, we develop a probabilistic model (PROLONG) for projecting global wind and solar power deployment. In our central projections, both wind and solar power grow similarly to Intergovernmental Panel on Climate Change 2 °C-compatible pathways and faster than in current policy scenarios. The COP28 pledge to triple renewables by 2030 is near the 95th percentile of our projections and requires that the growth of wind and solar photovoltaics in major economies accelerate by 1.4–3 times and 2–5 times, respectively. PROLONG can be adopted for data-driven projections of other policy-dependent energy technologies.

13:45-14:15 : Markus Pettersson, Chalmers — Beyond Point Predictions: Uncertainty-Aware Satellite Poverty Mapping for Public Policy

Household survey data underpin poverty policy but remain sparse across much of the developing world. Recent advances combining Earth observation imagery with machine learning have shown potential of filling in these data gaps, yet high-stakes decisions require more than accurate point estimates. While the average accuracy of these EO-ML predictions appears promising, we find that local uncertainty is substantial and often obscured by aggregate metrics. However, this does not render such models unusable. We show that explicitly incorporating uncertainty into decision rules, with a conformal prediction–based approach, can improve policy outcomes. Using these insights, we propose an aid-allocation strategy which reduces unnecessary survey costs while preserving statistical guarantees on targeting support.

14:15-14:45 : Romaric Duvignau, Chalmers — Algorithms for Peer-to-Peer Energy Sharing: From Small Communities to Scalable Systems
Peer-to-peer (P2P) energy sharing is a promising approach to better utilize distributed renewable generation and storage, but making such systems practical raises several algorithmic challenges, including how to form efficient communities and how to solve the resulting optimization problems at scale. In this talk, I present a set of works on P2P energy sharing developed within our research group over the past six years. Using real-world data, I show that most of the achievable cost savings can already be obtained with very small communities, which motivates simple and decentralized designs. I then discuss the main computational problems underlying these systems, in particular geographical peer matching and cost optimization, and present scalable algorithms that achieve a good trade-off between solution quality and computational cost. I will also briefly touch upon recent work on discovery algorithms for assignment problems, motivated by settings where relevant information is costly to obtain, and their connection to forming energy-sharing communities.

14:45-15:00 : Fika break

15:00-15:45 : Sonia Yeh, Chalmers — Toward Resilient and Equitable Mobility Systems with Dynamic Mobility Digital Twins

This project aims to develop new ways to represent and anticipate how people and mobility systems respond to disruptions such as heatwaves, wildfires, service outages, and power failures, using large-scale mobility data, generative AI, and agent-based simulation. By enabling policy and infrastructure stress-testing in a mobility digital twin, it seeks to support more resilient, equitable, and future-ready mobility planning.

15:00-15:45 : S. Keshav, Cambridge  — Democratizing Earth Observation with Foundation Models: The TESSERA Project and  Embedding Explorer

Foundation models for Earth observation (EO) promise to transform how we monitor and understand our planet, yet the computational cost of training and deploying such models risks concentrating their benefits among a small number of well-resourced institutions. In this talk, I present TESSERA, a foundation model for Earth observation that generates rich semantic embeddings from Sentinel-1 and Sentinel-2 imagery at 10-metre global resolution. I discuss the architectural and data engineering choices underpinning TESSERA, situating them within the rapidly evolving landscape of EO foundation models, and highlight the downstream tasks — from solar panel detection to habitat mapping — that these embeddings support. I will also discuss how TESSERA embeddings can be used for uncertainty-quantified estimates of geo-located point samples. 

A key ambition of the TESSERA project is to make state-of-the-art EO embeddings freely and practically accessible to researchers, policymakers, and conservation practitioners worldwide, including those without access to large-scale GPU infrastructure. To this end, I demonstrate the TESSERA Embedding Explorer (TEE), a lightweight, browser-based tool that allows users to interactively visualize, query, and analyse TESSERA embeddings over any region of interest without requiring specialized hardware or software. I discuss how tools like TEE can lower the barrier to entry for sophisticated geospatial analysis and reflect on the broader challenge of operationalizing foundation model outputs for real-world environmental decision-making.

16:30-17:00 : Open Discussion

The program can still be subject to small changes before its final version. 

1. How do I find the room?
   The room is EC in the EDIT building and directions can be found at this [link]

2. Do I need to register?
   Registration is not compulsory. Non-registered participants will be admitted until room capacity is reached and won't receive lunch.

3. I am not able to attend, can I join remotely?
   We are considering a hybrid format for the colloquium part of the event. Links will be provided as we get closer to the event's date.