Consistent Training via Energy-Based GFlowNets for Modeling Discrete Joint Distributions

Generative Flow Networks (GFlowNets) have demonstrated significant performance improvements for generating diverse discrete objects $x$ given a reward function $R(x)$, indicating the utility of the object and trained independently from the GFlowNet by supervised learning to predict a desirable property $y$ given $x$. We hypothesize that this can lead to incompatibility between the inductive optimization biases in training $R$ and in training the GFlowNet, potentially leading to worse samples and slow adaptation to changes in the distribution. In this work, we build upon recent work on jointly learning energy-based models with GFlowNets and extend it to learn the joint over multiple variables, which we call Joint Energy-Based GFlowNets (JEBGFNs), such as peptide sequences and their antimicrobial activity. Joint learning of the energy-based model, used as a reward for the GFlowNet, can resolve the issues of incompatibility since both the reward function $R$ and the GFlowNet sampler are trained jointly. We find that this joint training or joint energy-based formulation leads to significant improvements in generating anti-microbial peptides. As the training sequences arose out of evolutionary or artificial selection for high antibiotic activity, there is presumably some structure in the distribution of sequences that reveals information about the antibiotic activity. This results in an advantage to modeling their joint generatively vs. pure discriminative modeling. We also evaluate JEBGFN in an active learning setting for discovering anti-microbial peptides.

[1]  Bonaventure F. P. Dossou,et al.  Biological Sequence Design with GFlowNets , 2022, ICML.

[2]  S. Levine,et al.  Design-Bench: Benchmarks for Data-Driven Offline Model-Based Optimization , 2022, ICML.

[3]  Aaron C. Courville,et al.  Generative Flow Networks for Discrete Probabilistic Modeling , 2022, ICML.

[4]  Chen Sun,et al.  Trajectory Balance: Improved Credit Assignment in GFlowNets , 2022, NeurIPS.

[5]  Shuang Li,et al.  Active Learning for Domain Adaptation: An Energy-based Approach , 2021, AAAI.

[6]  A. Torralba,et al.  Energy-Based Models for Continual Learning , 2020, CoLLAs.

[7]  M. Barber,et al.  Diversity , 2010, The Fairchild Books Dictionary of Fashion.

[8]  Edward J. Hu,et al.  GFlowNet Foundations , 2021, J. Mach. Learn. Res..

[9]  Anima Anandkumar,et al.  Controllable and Compositional Generation with Latent-Space Energy-Based Models , 2021, NeurIPS.

[10]  Tingjun Hou,et al.  Multi-constraint molecular generation based on conditional transformer, knowledge distillation and reinforcement learning , 2021, Nature Machine Intelligence.

[11]  Richard Zemel,et al.  Directly Training Joint Energy-Based Models for Conditional Synthesis and Calibrated Prediction of Multi-Attribute Data , 2021, ArXiv.

[12]  Doina Precup,et al.  Flow Network based Generative Models for Non-Iterative Diverse Candidate Generation , 2021, NeurIPS.

[13]  Safwan Wshah,et al.  AMPGAN v2: Machine Learning-Guided Design of Antimicrobial Peptides , 2021, J. Chem. Inf. Model..

[14]  Yi Yan Yang,et al.  Accelerated antimicrobial discovery via deep generative models and molecular dynamics simulations , 2021, Nature Biomedical Engineering.

[15]  Bryan McCann,et al.  Joint Energy-based Model Training for Better Calibrated Natural Language Understanding Models , 2021, EACL.

[16]  Diederik P. Kingma,et al.  How to Train Your Energy-Based Models , 2021, ArXiv.

[17]  Alex Rosenthal,et al.  DBAASP v3: database of antimicrobial/cytotoxic activity and structure of peptides as a resource for development of new therapeutics , 2020, Nucleic Acids Res..

[18]  Alexandros Kalousis,et al.  Goal-directed Generation of Discrete Structures with Conditional Generative Models , 2020, NeurIPS.

[19]  Paul Rayson,et al.  BOSS: Bayesian Optimization over String Spaces , 2020, NeurIPS.

[20]  David Dohan,et al.  Model-based reinforcement learning for biological sequence design , 2020, ICLR.

[21]  Aleksandra Mojsilovic,et al.  CogMol: Target-Specific and Selective Drug Design for COVID-19 Using Deep Generative Models , 2020, NeurIPS.

[22]  Yoshua Bengio,et al.  Your GAN is Secretly an Energy-based Model and You Should use Discriminator Driven Latent Sampling , 2020, NeurIPS.

[23]  B. Baker,et al.  Diversity, Ecology, and Prevalence of Antimicrobials in Nature , 2019, Front. Microbiol..

[24]  Liang Lin,et al.  Multivariate-Information Adversarial Ensemble for Scalable Joint Distribution Matching , 2019, ICML.

[25]  Igor Mordatch,et al.  Implicit Generation and Generalization with Energy Based Models , 2018 .

[26]  Kyunghyun Cho,et al.  Conditional molecular design with deep generative models , 2018, J. Chem. Inf. Model..

[27]  Guoyin Wang,et al.  JointGAN: Multi-Domain Joint Distribution Learning with Generative Adversarial Nets , 2018, ICML.

[28]  L. Deng,et al.  The MNIST Database of Handwritten Digit Images for Machine Learning Research [Best of the Web] , 2012, IEEE Signal Processing Magazine.

[29]  Fu Jie Huang,et al.  A Tutorial on Energy-Based Learning , 2006 .

[30]  Miguel Á. Carreira-Perpiñán,et al.  On Contrastive Divergence Learning , 2005, AISTATS.