Research Statement

Toward a theory of
ethical embeddings

Three problems that look separate — who a recommender makes visible, what a language model associates with whom, and what a sanitized text still means — are the same problem in different clothes. Each is a question about what a representation encodes, and who pays for it.

01 / Fairness in recommendation

Exposure is a resource, and it is allocated

Recommendation algorithms decide visibility. MinWalk optimizes for accuracy and equitable exposure together, integrating exposure-parity constraints with dynamic network modelling rather than treating fairness as a post-hoc correction.

  • Mitigates homophily and popularity bias, raising minority visibility over time.
  • Constrained optimization with stochastic sampling, so it scales to large graphs.
  • On social and citation graphs, visibility gains for under-represented users at minimal precision cost.
Bar charts comparing MinWalk against baselines across the congress, email_eu, wiki_vote and facebook datasets

MinWalk vs baselines across four network datasets. Click to enlarge.

Stacked bar charts showing gender-linked association across BGE, OpenAI, Cohere, Google and Microsoft embedding models by top-k words

Gender-linked association across five providers, by top-k words. Click to enlarge.

02 / Bias in language models

A bias you cannot measure is one you cannot fix

I quantify demographic bias in embeddings from major providers using frequency-weighted Fightin' Words, SC-WEAT, clustering, and divergence tests. The output is not an opinion about a model. It is a number, with a method behind it.

  • Interpretable scores expose gender and race associations that survive into downstream tasks, including product recommendation.
  • Multi-method audit pipelines (Marked Words, SVM, Jensen–Shannon divergence) that point at actionable mitigations.
  • Audited across OpenAI, Google, Microsoft, Cohere, and BGE.
03 / Privacy-preserving NLP

Protect the person, keep the meaning

I work on differentially private text by perturbing the embedding space rather than token probabilities. That preserves coherence while reducing memorization and leakage, which is the trade every deployment actually faces.

  • CluSanT — sanitization under Metric Local Differential Privacy, with clustering to hold semantics at a given budget.
  • Lower identifiable memorization while retaining high human-rated quality.
  • Applications in healthcare narratives, legal anonymization, and educational feedback.
Two heatmaps comparing CluSanT against CusText at privacy budgets epsilon = 1 and epsilon = 8

CluSanT vs CusText — semantic similarity gain at ε = 1 and ε = 8. Click to enlarge.

Next five years

Where this goes

The three threads converge. Each is a claim about what a representation encodes; the goal is one framework that treats them as such.

  • Dynamic fairness

    Treat fairness as a temporal property of evolving networks, not a snapshot. Sustain minority visibility at scale, over time.

  • Systemic bias

    Move past local prompt patches toward structural rebalancing of embedding spaces themselves.

  • Privacy as transformation

    Coherence-preserving, embedding-level privacy integrated directly into the generation loop.

  • Ethical embeddings

    A unified mathematical framework connecting visibility, stereotype, and attribute leakage as one object.

Method

The toolkit

Cross-domain and reproducible, spanning NLP and recommender systems, and built to hold up in compliance contexts such as the EU AI Act and the NIST AI Risk Management Framework.

Embedding analysis · PCA, cosine SC-WEAT & directional tests Exposure & demographic parity Disparity ratios · equal opportunity Custom losses · Lagrangian relaxation DP-SGD · embedding perturbation BLEU · ROUGE · perplexity · CoLA BERT · RoBERTa · GPT · Cohere · BGE
Contact

Read the full statement

The complete research statement goes deeper on methods, results, and the five-year programme.

Vancouver, British Columbia
Division of Infectious Diseases, UBC

← Back to the main site