Interactive notebook
Explore My Research
Start with a question, not a publication list: pick an interest and a reading pace, then follow a route through the papers.
Runs in your browser: nothing is saved and no account is needed. Optional AI features and Paper Q&A show their own notice and send nothing until you submit.
02Follow the connections
Your reading route
Choose a topic or describe a question to begin.
Browse all papers without a route
- Mechanism Design for AI Overviews: Creator Incentives and Long-Term Profit
What happens to costly content creation when AI search redirects traffic? This paper studies citation and compensation mechanisms and long-term platform profit.
- From Seller to Data Supplier: How Creator Compensation Enables AI Displacement
This unpublished working draft studies how paying for training data can let AI replace creators' direct sales while retaining them as data suppliers. Its model distinguishes displacement from creator welfare and compares compensation with uncompensated data provision.
- How Sampling Shapes LLM Alignment: From One-Shot Optima to Iterative Dynamics
How do sampling and reference policies change preference alignment when a learned policy feeds back into the next round? The paper studies ranking, concentration, oscillation, and stability under specified regimes.
- Human vs Generative AI in Content Creation Competition: Symbiosis or Conflict?
A competition model generalized from a Tullock contest examines human and generative-AI content creation, including conditions supporting coexistence.
- How Bad is Top-K Recommendation under Competing Content Creators?
When creators compete for attention, how much user welfare can top-K recommendation lose? This paper analyzes welfare under random user choices and no-regret creator learning.
- Policy Design for Two-sided Platforms with Participation Dynamics
Viewers and providers affect each other's participation. This paper studies policies that account for these population effects rather than optimizing only immediate outcomes.
- User Welfare Optimization in Recommender Systems with Competing Content Creators
How can a platform use its information about users to influence competing creators? This work studies welfare-oriented recommendation and reward interventions, with offline and online experiments.
- The Complexity of Tullock Contests
This paper studies when pure-equilibrium computation in heterogeneous Tullock contests is tractable, and how elasticity affects the computational problem.
- Multi-Agent Learning for Iterative Dominance Elimination: Formal Barriers and New Algorithms
Can multi-agent bandit learning efficiently eliminate dominated actions? This work identifies barriers for standard regret-based algorithms and studies Exp3-DH.
- Single-Agent Poisoning Attacks Suffice to Ruin Multi-Agent Learning
The paper examines utility-observation poisoning in strongly monotone games and the trade-off between learning efficiency and robustness.
- CREDIT: Certified Ownership Verification of Deep Neural Networks Against Model Extraction Attacks
CREDIT studies certified ownership verification against model extraction, using mutual information and a verification threshold.
- PAC-Learning for Strategic Classification
Strategic VC-dimension extends a PAC-learning framework to people with heterogeneous incentives to manipulate their testing data.
Descriptions are editorial reading aids based on the publication summaries; check the source for precise assumptions and claims. Some working drafts are not downloadable; use their paper assistants for conceptual questions.