Research map / Probabilistic and scientific modeling / Molecules and proteins
Data-Driven Knowledge-Aware Inference of Private Information in Continuous Double Auctions
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes KATE, a knowledge-aware neural framework that infers sellers' private cost vectors in multi-unit continuous double auctions by training a GRU behavior model with a loss that matches model-based posteriors to knowledge-based posteriors derived from a shifted-exponential ask-cost ratio distribution.
Problem
Inferring human sellers' private cost vectors from their observed asks in continuous double auctions, where equilibria are intractable and per-individual private costs are unobserved.
Approach
Learn a neural behavior function by penalizing divergence between model-based posteriors over private costs (from Bayes' rule on the behavior model) and knowledge-based posteriors derived from the population distribution of ask-cost ratios, combined with cross-entropy and marginal-cost monotonicity losses.
Key results
- KATE achieves the lowest test MSE of about 887.89, versus 2240.37 for the best knowledge-unaware baseline (BLUE).
- Among knowledge-aware methods, KATE (887.89) outperforms DL-KL (1116.60), SL-KL (1100.30), and DI-KP (1284.03).
- KATE reduces MSE by about 410 compared with the best feasible-set baseline (Median, 1298.62).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use KATE instead of BLUE | Private cost inference in CDAs · Testing data | MSE: 2240.37 -> 887.89 | Helped |
| Use KATE instead of CE | Private cost inference in CDAs · Testing data | MSE: 2992.16 -> 887.89 | Helped |
| Use KATE instead of DL | Private cost inference in CDAs · Testing data | MSE: 6215.68 -> 887.89 | Helped |
| Use KATE instead of SL | Private cost inference in CDAs · Testing data | MSE: 6213.76 -> 887.89 | Helped |
| Use KATE instead of DL-KL | Private cost inference in CDAs · Testing data | MSE: 1116.60 -> 887.89 | Helped |
| Use KATE instead of SL-KL | Private cost inference in CDAs · Testing data | MSE: 1100.30 -> 887.89 | Helped |
| Use KATE instead of DI-KP | Private cost inference in CDAs · Testing data | MSE: 1284.03 -> 887.89 | Helped |
| Use KATE instead of Median | Private cost inference in CDAs · Testing data | MSE: 1298.62 -> 887.89 | Helped |
| Use KATE instead of Mean | Private cost inference in CDAs · Testing data | MSE: 1333.91 -> 887.89 | Helped |
| Use KATE instead of Uniform | Private cost inference in CDAs · Testing data | MSE: 2624.50 -> 887.89 | Helped |
| Use KATE instead of Truthfulness-Based Inference | Private cost inference in CDAs · Testing data | MSE: 5064.17 -> 887.89 | Helped |
| Use knowledge of ask-cost ratios in learning-based inference | Private cost inference in CDAs · Testing data | MSE: 2240.37 -> 887.89 | Helped |
Surprising findings
- Direct inference from the knowledge-based posterior alone (DI-KP) is the least accurate knowledge-aware method, showing that the learned behavior function is needed to exploit prior offers.
- A simple Median-of-feasible-set estimator beats the Uniform and truthfulness-based baselines, because most costs lie near the medians of their feasible intervals in this dataset.
Limitations
- Evaluated on a single experimental classroom dataset (MobLab) with costs ranging 1 to 300 and vector size up to 3.
- Median and Mean performance may deteriorate on datasets lacking the feature that costs cluster near feasible-interval medians.
- Assumes the ask-cost ratio follows a shifted exponential distribution, which was validated only on this dataset.
Experimental setup
Datasets: MobLab CDA experimental dataset (Lin et al. 2020)
Benchmarks: MobLab CDA experimental dataset (Lin et al. 2020)
Metrics: mean squared error
Baselines: CE BLUE DL SL DL-KL SL-KL DI-KP Median Mean Uniform Truthfulness-based inference
Builds on: BLUE (Cui and Yu 2023) dual learning inverse game theory
Closest papers by meaning
- Increasing Revenue in Efficient Combinatorial Auctions by Learning to Generate Artificial Competition (AAAI 2025 · similarity 0.85)
- Predicting Real-World Penny Auction Durations by Integrating Game Theory and Machine Learning (AAAI 2024 · similarity 0.84)
- Safely Learning Optimal Auctions: A Testable Learning Framework for Mechanism Design (ICML 2025 · similarity 0.84)
- Beyond Monotonicity: On the Convergence of Learning Algorithms in Standard Auction Games (AAAI 2025 · similarity 0.83)
- A Parametric Contextual Online Learning Theory of Brokerage (ICML 2025 · similarity 0.83)
- Feature-Based Online Bilateral Trade (ICLR 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.