latent representation
Latent representations are abstract features learned by a model, often through techniques like autoencoders, that capture essential characteristics of the data without the noise. These representations are beneficial for tasks like clustering, generation, and downstream predictive tasks.
- COLA: Towards Efficient Multi-Objective Reinforcement Learning with Conflict Objective Regularization in Latent Space
- Continuous Thought Machines
- Disentangled Representation Learning via Modular Compositional Bias
- FreeInv: Free Lunch for Improving DDIM Inversion
- Glocal Information Bottleneck for Time Series Imputation
- MaNGO — Adaptable Graph Network Simulators via Meta-Learning
- Measuring and Guiding Monosemanticity
- RepLDM: Reprogramming Pretrained Latent Diffusion Models for High-Quality, High-Efficiency, High-Resolution Image Generation
- State-Covering Trajectory Stitching for Diffusion Planners
- UMAMI: Unifying Masked Autoregressive Models and Deterministic Rendering for View Synthesis
- Unmasking Puppeteers: Leveraging Biometric Leakage to Expose Impersonation in AI-Based Videoconferencing