The impact of training method on odour learning and generalisation in detection animals

Rats can distinguish (and generalize) among two white wine varieties

In the olfactory literature there is considerable debate about how differences in olfactory receptors across different species map onto variations in perceptual acuity and performance. Although humans have fewer functional olfactory receptors than most other mammals, it has been suggested that linguistic and cognitive abilities help compensate for this apparent deficit and enhance discriminative abilities, particularly through humans' ability to categorize sensory stimuli into conceptual categories. However, previous research suggests that non-human animals can learn complex categories, involving multiple perceptual dimensions, indicating that they can discriminate complex odor stimuli without language. We investigated generalization over complex olfactory categories by examining rats' discrimination of wine varieties, a challenging task for humans that has been suggested to rely heavily on human-specific linguistic, cognitive and categorization abilities. Nine rats were trained in an olfactory discrimination task (go/no-go) using a specific wine variety (Riesling or Sauvignon Blanc from different winemakers) as the S + . Rats were then tested using novel wines of the same varieties in unrewarded probe trials to assess their abilities to correctly assign instances of wine to specific categories. Interestingly, all nine rats successfully learned to discriminate the two varieties, and most rats generalized within two test trials to novel wines of the same varieties. We explore the implications of our results for olfactory concept formation and categorization more generally.

Odour generalisation and detection dog training

Detection dogs are required to search for and alert to specific odours of interest, such as drugs, cadavers, disease markers and explosives. However, the odour released from different samples of the same target substance will vary for a number of reasons, including the production method, evaporation, degradation, or by being mixed with extraneous odours. Generalisation, the tendency to respond in the same manner to stimuli which are different - but similar to - a conditioned stimulus, is therefore a crucial requirement for working detection dogs. Odour is a complex modality which poses unique challenges in terms of reliably predicting generalisation, when compared with auditory or visual stimuli. The primary aim of this review is to explore recent advances in our understanding of generalisation and the factors that influence it, and to consider these in light of detection dog training methods currently used in the field. We identify potential risks associated with certain training practices, and highlight areas where research is lacking and which warrant further investigation.

Evaluation of canine detection of covid-19 infected individuals under controlled settings

Reverse transcription polymerase chain reaction (RT‐PCR) is currently the standard diagnostic method to detect symptomatic and asymptomatic individuals infected with Severe acute respiratory syndrome coronavirus 2 (SARS‐CoV‐2). However, RT‐PCR results are not immediate and may falsely be negative before an infected individual sheds viral particles in the upper airways where swabs are collected. Infected individuals emit volatile organic compounds in their breath and sweat that are detectable by trained dogs. Here, we evaluate the diagnostic accuracy of dog detection against SARS‐CoV‐2 infection. Fifteen dogs previously trained at two centres in Australia were presented to axillary sweat specimens collected from known SARS‐CoV‐2 human cases (n = 100) and non‐cases (n = 414). The true infection status of the cases and non‐cases were confirmed based on RT‐PCR results as well as clinical presentation. Across dogs, the overall diagnostic sensitivity (DSe) was 95.3% (95%CI: 93.1–97.6%) and diagnostic specificity (DSp) was 97.1% (95%CI: 90.7–100.0%). The DSp decreased significantly when non‐case specimens were collected over 1 min rather than 20 min (p value = .004). The location of evaluation did not impact the detection performances. The accuracy of detection varied across dogs and experienced dogs revealed a marginally better DSp (p value = .016). The potential and limitations of this alternative detection tool are discussed.

Case Study: An Evaluation of Detection Dog Generalization to a Large Quantity of an Unknown Explosive in the Field

Simple Summary

This case study is a series of experiments to investigate a real-life event where two highly trained and certified detection dogs did not find an explosive in a suspicious bag. We tested seven dog teams from the agency in three experiments and confirmed that dogs were able to detect the agency’s training sample in a small quantity (30 g) but not the large amount of the confiscated explosive (13 kg) found in a similar scenario. To further evaluate a possible generalization deficit, we tested dogs with a 30 g subsample of the confiscated explosive, and most of the dogs were able to detect it (but with some decrement) even though they largely failed to detect 13 kg of the same material. Finally, we trained dogs to detect the 30 g subsample until reaching proficiency and found that after training with the small sample, dogs still showed poor generalization to the large-quantity sample until explicit training with the large sample was conducted. Altogether, this series of studies shows the importance of explicitly training for larger quantity finds and has led to changes in agency practices to mitigate future risks.

Abstract

Two explosive detection dogs were deployed to search a suspicious bag, and failed to detect 13 kg of explosive within. The aim of this research was to further evaluate this incident. First, dog teams (N = 7) searched four bags in a similar scenario. One bag contained the same 13 kg of explosive, two bags were blanks, and the other contained the training sample that the agency routinely used for training. All dogs detected the training sample, but most (5/7) did not alert to the 13 kg sample. Subsequently, dogs received two trials in a line up with a 30 g subsample of the explosive to evaluate whether they could generalize to a smaller quantity. Most dogs (6/7) alerted to the subsample at least once. Finally, dogs were trained with the 30 g subsample and later tested with the 13 kg sample. Only three dogs spontaneously generalized to the large sample after training with the small subsample. Dogs’ alert rate to the 13 kg sample was improved with training in subsequent trials with the 13 kg sample. This result indicates that explosive detection dogs may not generalize to a target odor at a significantly higher quantity relative to the one used in training, highlighting the importance of conducting such training.