Figure 3 - available via license: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International
Content may be subject to copyright.
Source publication
Neurosymbolic Programming (NP) techniques have the potential to accelerate scientific discovery across fields. These models combine neural and symbolic components to learn complex patterns and representations from data, using high-level concepts or known constraints. As a result, NP techniques can interface with symbolic domain knowledge from scien...
Contexts in source publication
Context 1
... it is difficult to diagnose errors and interpret the model output. NP models have the potential to produce symbolic descriptions of behavior (Figure 3), which enables experts to connect model interpretations with other parts of the behavior analysis workflow. For example, for describing behavioral differences across different strains of mice. ...
Context 2
... et al., 2020] claim that programs synthesized by NEAR compete in performance with a black-box recurrent neural network while being much more human-interpretable. Figure 3 is an example program discovered by NEAR for behavior analysis. ...
Citations
... 11,14 We will discuss research in the direction of scientist-in-the-loop AI systems toward tackling these challenges. 5,15,16 These systems are developed in collaboration with neuroscientists and computer scientists, in order to transform diverse and complex experimental data into interpretable representations. ...
... Second, we need human-interpretable models to enable bidirectional knowledge transfer. This is an active area with works in explainable AI and machine learning, 46 mechanistic interpretability,47 and neurosymbolic modeling.16,48 Further research in these areas enables veterinarians to both integrate their domain expertise into the model and extract new insights and knowledge from the model's outputs. ...
This review focuses on opportunities and challenges of future AI developments in veterinary medicine, from the perspective of computer science researchers in developing AI systems for animal behavior analysis. We examine the paradigms of supervised learning, self-supervised learning, and foundation models, highlighting their applications and limitations in automating animal behavior analysis. These emerging technologies present future challenges in data, modeling, and evaluation in veterinary medicine. To address this, we advocate for a collaborative approach that integrates the expertise of AI researchers, veterinary professionals, and other stakeholders to navigate the evolving landscape of AI in veterinary medicine. Through cross-domain dialogue and an emphasis on human and animal well-being, we can shape AI development to advance veterinary practice for the benefit of all.