Machine Learning for Tangible Effects: Natural Language Processing for Uncovering the Illicit Massage Industry & Computer Vision for Tactile Sensing
Bibliographic record
Abstract
I explore two questions in this thesis: how can computer science be used to fight human trafficking? And how can computer vision create a sense of touch? The United States illicit massage industry (IMI) is a multi-billion dollar industry that offers not just therapeutic massages but also commercial sexual services. Illicit massage parlors number in the thousands and exist in every major city in the United States. Employees are often immigrant women with few other job opportunities, leaving them vulnerable to fraud, coercion, and other facets of human trafficking. By creating datasets using three publicly-accessible websites: Google Places, Rubmaps, and the AMPReviews forum, I show how we can use natural language processing tools such as bag-of-words combined with machine learning classifiers to help monitor spatiotemporal trends in the IMI. Monitoring plays an essential role in preventing trafficking and protecting employees within the IMI. I further show how to use word embeddings such as Word2Vec to derive insights into the labor pressures and language barriers affecting IMI employees. Similarly, I analyze the income, demographics, and societal pressures (such as relationship status) affecting sex buyers. Other insights include linked domains and using the word embeddings as a tool for acronym expansion. I also consider counter-trafficking in the banking sector. Human trafficking is about money, much of which will eventually flow through the legal financial system. Banks are legally required to have safeguards to guarantee they are not aiding or abetting criminal activity. My preliminary work focuses on creating synthetic transaction data so that researchers can more easily prototype, evaluate, and collaborate on developing anti-money laundering algorithms. This work adopts agent-based modeling and is inspired by red-flagged transactional behaviors from the United States and the Canadian financial regulatory agencies. I show both the uses and limitations of my model in generating timestamps and payee-recipient graphs for transactions. Finally, I consider the role of computer vision in creating tactile sensors. Tactile sensors are critical for robots that seek to manipulate and interact with the world, a prerequisite for helping with household tasks. Existing sensors include the Gelsight sensor, which consists of a camera facing a gel that is lit from multiple angles. The surface of the gel is slightly translucent and slightly reflective (semi-specular), and when objects are pressed into the gel, the image becomes a tactile image. Adapting a Gelsight sensor to the task of finding buried objects in sand required several modifications. Creating a wedge-shaped sensor allows for digging down into the granular media. The novel use of fluorescent paint instead of LEDs for gel lighting allows for significant sensor size reduction. Finally, an integrated vibrator motor counteracts jamming in the granular media, reducing force requirements for moving through the media. This work also shows how to use a webcam and a printed reference marker, or fiducial, to create a low-cost six-axis force-torque sensor. Commercial six-axis force-torque sensors cost thousands of dollars and often contain delicate strain gauges. By contrast, this sensor is inexpensive, made using readily-available rapid prototyping technologies, and easy to modify. All code and hardware design files are open sourced, opening up six-axis force-torque sensing to a wider range of applications.
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.
How this classification was reachedexpand
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 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.002 | 0.000 |
| Scholarly communication | 0.002 | 0.005 |
| Open science | 0.001 | 0.002 |
| 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 itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".