Celeba
PulseAugur coverage of Celeba — every cluster mentioning Celeba across labs, papers, and developer communities, ranked by signal.
7 day(s) with sentiment data
-
New framework exposes privacy risks in federated LoRA-adapted Vision Transformers
Researchers have developed a new framework to address privacy risks in federated learning when using parameter-efficient fine-tuning methods like LoRA with Vision Transformers. The proposed spatially anchored generative…
-
Generated images remain statistically distinguishable from real ones, study finds
Researchers have investigated the statistical indistinguishability between real and generated images, finding that even with progressive abstractions like blurring or downsizing, classifiers can still differentiate betw…
-
New LOGSI method enhances guidance for generative models using Stochastic Interpolants
Researchers have developed LOGSI, a novel method for guiding generative models that utilize Stochastic Interpolants (SI) and Flow Matching (FM). Unlike previous approaches that focused on conditional scores, LOGSI addre…
-
New framework CRLoMTL improves multi-task learning with low-rank matrices
Researchers have developed CRLoMTL, a novel framework designed to improve multi-task learning (MTL) in computer vision by addressing negative transfer. This method utilizes low-rank matrices to resolve conflicts between…
-
New AI frameworks tackle complex data representation and panoramic reconstruction
Researchers have introduced Variational Atlas Autoencoders (VAA), a novel framework for representing complex, high-dimensional data by learning an atlas of local latent representations. This approach addresses the limit…
-
New Part-DINO framework enhances image part discovery using structural cues
Researchers have developed Part-DINO, a novel framework designed to improve part discovery in images by integrating structural information with self-supervised features from Vision Transformers (ViTs) like DINO. Existin…
-
Research paper analyzes optimal multitask representations under information constraints
A new research paper analyzes the optimal pretraining objectives for multitask representations under an information budget. The study formalizes this as a multitask information bottleneck, using a mutual information con…
-
New framework formalizes compositional reasoning in vision-language models
Researchers have introduced a new framework called Geodesically Decomposable Embeddings (GDE) that formalizes how large vision-language models encode compositional semantic structures. This framework identifies directio…
-
New dataset AfriGuard enhances LLM safety for 10 African languages
Researchers have developed AfriGuard, a new dataset and set of guardrail models designed to improve the safety and refusal capabilities of large language models (LLMs) for 10 African languages and English. Current LLMs …
-
New research questions linear probe methodology in vision models
A new paper questions the common methodology of using linear probes to determine if pretrained vision models encode generative factors linearly. The research suggests that high probe accuracy can be achieved by models w…
-
New CS-VGB method improves generative model sampling accuracy
Researchers have developed a new inference-time scaling method called Continuous-State Verifier-Guided Backtracking (CS-VGB) for continuous generative models. This technique aims to improve sampling accuracy by using ad…
-
New 'Hide and Seek' method explains black-box vision models without weights
Researchers have developed a new method called Hide and Seek (HnS) for interpreting black-box vision models. Unlike previous methods, HnS does not require access to model weights or concept-annotated datasets. It works …
-
New linear programming approach prunes datasets without labels
Researchers have developed a novel method for dataset pruning that uses linear programming to select a representative subset of data without requiring labels or model training. This approach reformulates unbiased subset…
-
New EOFlows method advances unsupervised feature discovery in AI
Researchers have introduced Entropy-Ordered Flows (EOFlows), a novel framework for unsupervised feature discovery in representation learning. This method utilizes a normalizing flow augmented with an orthogonality regul…
-
New research advances diffusion language models with efficiency and performance gains · 10 sources tracked
Researchers are developing new methods to improve diffusion language models (DLMs), focusing on efficiency and performance. "Clock Diffusion" introduces semi-autoregressive continuous DLMs with position-dependent noise …
-
CyclOT framework learns quadratic optimal transport maps from unpaired data
Researchers have introduced CyclOT, a novel neural framework for learning quadratic optimal transport maps from unpaired samples in high dimensions. This bidirectional approach utilizes synchronized forward-backward int…
-
New method LEAPSC enhances privacy in deep joint source-channel coding
Researchers have developed LEAPSC, a novel method for privacy-preserving deep joint source-channel coding. This technique integrates in-loop least-squares concept erasure within a variational information bottleneck enco…
-
New SAGE framework tackles spurious correlations in ML models
Researchers have developed SAGE, a novel two-stage generative augmentation framework designed to address spurious correlations in machine learning models. This approach uses cluster-derived sub-labels and class labels t…
-
New distillation method speeds up face synthesis by over 4x
Researchers have developed a method to accelerate the generation of synthetic face images using diffusion models. By distilling knowledge from the Arc2Face model into a latent Consistency Model, they achieved a 4.36x sp…
-
Text-guided models show bias in facial editing, study finds
A new study published on arXiv evaluates six text-guided diffusion models for facial editing tasks, comparing their performance against established methods like GANs and 3DMMs. The research introduces Face-Edit-Attribut…