Learning generative models from a control perspective
Bibliographic record
Abstract
The main topic of this thesis is to bridge two important domains in machine learning: probabilistic generative models and control methods. Generative models are used for modeling distributions of various kinds of data and are of great importance in creative generation applications. Generative models can also be used for versatile probabilistic inference tasks, such as posterior solving or normalizing factor estimation, which are important in many scientific applications. The focus of this thesis is to study how control insights can guide us to design better generative modeling algorithms, for example, achieving a better trade-off between exploration and exploitation. We present a series of our works on this aspect. In the first work, we propose energy-based GFlowNet to jointly train a GFlowNet as well as an energy-based model for modeling discrete data. For continuous distributions, we investigate how to train a hierarchical variational model as a GFlowNet in a diffusion-like way. Lastly, we extend our approach to graph combinatorial optimization problems to sample high-quality solutions. Together, these contributions advance sample quality and steerability of generative models through new training objectives and inference techniques. Last but not least, we outline ongoing and future work from both the algorithmic perspective and the applicative angles in the last chapter.
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 | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.002 | 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 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".