Automated Emotion Classification in Free-moving Rats
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Studies involving emotion often use animal models and currently rely on manual labelling by researchers. This human-driven labelling approach leads to a number of challenges such as: long analysis times, imprecise results, observer drift, and varying correlation between observers. These problems impact reproducibility, and have contributed to our lack of understanding of fundamental mechanical questions such as how emotions arise from neuronal circuits. Recent success of machine learning models across similar problems show that it can help to mitigate these challenges while meeting or exceeding human accuracy. 
 We developed a classifier pipeline that takes in videos and produces an emotion label. The pipeline extracts body part positions from each frame using a pose estimator and feeds them into an Artificial Neural Network (ANN) classifier built using stacked Long Short Term Memory (LSTM) layers. The data was collected by treating nine rats with Lypopolysaccharide (LPS) injections (10mg/kg). First, rats were recorded for 10 minutes under control conditions with no manipulation and no observed symptoms of stress or malaise. A week later, rats were injected with LPS and filmed for 10 minutes two hours post-injection. 
 The classifier pipeline developed correctly labelled 78% of the 125,040 video segments from 8 test videos. When combined with a vote-based system, this led to 7 of the 8 test videos being classified correctly which was the same accuracy attained by a human expert from the lab. The test videos had varying environments and used rats that were different from the training videos, providing evidence of a degree of robustness in the model. Future work will focus on expanding the test data and incorporating models for 3D pose estimation and behavioral classification.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it