Cold decoding: Energy-based constrained text generation with langevin dynamics

L Qin, S Welleck, D Khashabi… - Advances in Neural …, 2022 - proceedings.neurips.cc
Many applications of text generation require incorporating different constraints to control the
semantics or style of generated text. These constraints can be hard (eg, ensuring certain …

Latent space energy-based model for fine-grained open set recognition

W Bao, Q Yu, Y Kong - arxiv preprint arxiv:2309.10711, 2023 - arxiv.org
Fine-grained open-set recognition (FineOSR) aims to recognize images belonging to
classes with subtle appearance differences while rejecting images of unknown classes. A …

Gedi: Generative and discriminative training for self-supervised learning

E Sansone, R Manhaeve - arxiv preprint arxiv:2212.13425, 2022 - arxiv.org
Self-supervised learning is a popular and powerful method for utilizing large amounts of
unlabeled data, for which a wide variety of training objectives have been proposed in the …

Learning symbolic representations through joint generative and discriminative training

E Sansone, R Manhaeve - arxiv preprint arxiv:2304.11357, 2023 - arxiv.org
We introduce GEDI, a Bayesian framework that combines existing self-supervised learning
objectives with likelihood-based generative models. This framework leverages the benefits …

Lazy Layers to Make Fine-Tuned Diffusion Models More Traceable

H Liu, W Zhang, B Li, B Ghanem… - arxiv preprint arxiv …, 2024 - arxiv.org
Foundational generative models should be traceable to protect their owners and facilitate
safety regulation. To achieve this, traditional approaches embed identifiers based on …

Maximum entropy exploration in contextual bandits with neural networks and energy based models

A Elwood, M Leonardi, A Mohamed, A Rozza - Entropy, 2023 - mdpi.com
Contextual bandits can solve a huge range of real-world problems. However, current
popular algorithms to solve them either rely on linear models or unreliable uncertainty …

A Bayesian Unification of Self-Supervised Clustering and Energy-Based Models

E Sansone, R Manhaeve - arxiv preprint arxiv:2401.00873, 2023 - arxiv.org
Self-supervised learning is a popular and powerful method for utilizing large amounts of
unlabeled data, for which a wide variety of training objectives have been proposed in the …

The Triad of Failure Modes and a Possible Way Out

E Sansone - arxiv preprint arxiv:2309.15420, 2023 - arxiv.org
We present a novel objective function for cluster-based self-supervised learning (SSL) that is
designed to circumvent the triad of failure modes, namely representation collapse, cluster …

Triple-Hybrid Energy-based Model Makes Better Calibrated Natural Language Understanding Models

H Xu, Y Zhang - Proceedings of the 17th Conference of the …, 2023 - aclanthology.org
Though pre-trained language models achieve notable success in many applications, it's
usually controversial for over-confident predictions. Specifically, the in-distribution (ID) …

[PDF][PDF] Unifying Self-Supervised Clustering and Energy-Based Models

E Sansone, R Manhaeve - arxiv preprint arxiv:2401.00873, 2024 - researchgate.net
Self-supervised learning excels at learning representations from large amounts of data. At
the same time, generative models offer the complementary property of learning information …