Andrii Krutsylo

This paper presents outcomes from the inaugural "EcoHack: AI & LLM Hackathon for Applications in Evidence-based Ecological Research & Practice," which convened participants from across Europe and beyond, culminating in 11 team submissions. These submissions highlighted six broad application areas of AI for ecology: (1) AI-enhanced decision support and automation, (2) scientific search and communication, (3) knowledge extraction and reasoning, (4) AI for ecological modeling, forecasting, and simulation, (5) causal inference and ecological reasoning, and (6) AI for biodiversity monitoring and conservation. Each team’s project is summarized in a consolidated table—complete with links to source code—and described in brief papers in the appendix. Beyond summarizing technical results, this paper offers insights into the hackathon’s hybrid structure, featuring an in-person gathering in Bielefeld, Germany, alongside a global online hub that facilitated both local and virtual engagement. Throughout the event, participants showcased how large language models (LLMs) can serve as both robust tools for diverse machine learning tasks and flexible platforms for rapidly prototyping novel research applications. These efforts underscore the importance of stronger technological bridges among stakeholders in ecology, including practitioners, local farmers, and policymakers. Overall, the EcoHack outcomes highlight the transformative potential of AI in driving scientific discovery and fostering interdisciplinary collaboration in ecology.

Learning continually from a stream of non-i.i.d. data is an open challenge in deep learning, even more so when working in resource-constrained environments such as embedded devices. Visual models that are continually updated through supervised learning are often prone to overfitting, catastrophic forgetting, and biased representations. On the other hand, large language models contain knowledge about multiple concepts and their relations, which can foster a more robust, informed and coherent learning process. This work proposes Continual Visual Mapping (CVM), an approach that continually ground vision representations to a knowledge space extracted from a fixed Language model. Specifically, CVM continually trains a small and efficient visual model to map its representations into a conceptual space established by a fixed Large Language Model. Due to their smaller nature, CVM can be used when directly adapting large visual pre-trained models is unfeasible due to computational or data constraints. CVM overcome state-of-the-art continual learning methods on five benchmarks and offers a promising avenue for addressing generalization capabilities in continual learning, even in computationally constrained devices.

Continual learning presents a machine learning paradigm in which models are expected to learn new tasks sequentially without forgetting previously acquired knowledge. Measuring the quality of this preserved knowledge remains a challenge, particularly when the model has access to a limited number of prior experiences and risks overfitting rather than enhancing generalization. To address this issue, we introduce a novel Knowledge Retention metric capable of determining whether a model retains useful information or merely optimizes for short-term performance. For memory-based continual learning approaches, the proposed metric helps identify if the evaluated method uses memory for re-learning the tasks from scratch, providing valuable information about both the memory and the method that uses it.

In a Continual Learning setting, models are trained on data with occasional distribution shifts, resulting in forgetting the information learned before each shift. Experience Replay (ER) addresses this challenge by retaining part of the old training samples and replaying them alongside current data, improving the model's understanding of the overall distribution in training batches. The crucial factor in ER performance is the diversity of samples within batches. The impact of sample diversity across a sequence of batches is investigated, introducing a new metric and an associated approach to assess and leverage this diversity. This exploration opens up significant potential for future work, as various strategies can be devised to ensure inter-batch diversity. Achieving optimal results may involve striking a balance between this novel metric and other inherent properties of a batch or sequence.