Posters

Researchers uncover hidden training data secrets of large language models by analyzing their byte-pair encoding tokenizers, revealing the proportions of different languages and domains.

Data-Free Backdoor Attacks (DFBA) injects undetectable backdoors into pre-trained classifiers without retraining or architectural changes, bypassing existing defenses.

This paper proposes a novel MMD-based method for data distribution valuation, enabling theoretically-principled comparison of data distributions from limited samples, outperforming existing methods in…

Boosting semi-supervised learning, a new data augmentation method using diffusion models significantly improves model accuracy, especially with mismatched data distributions.

Unlearning synthesized images efficiently reveals influential training data for text-to-image models, improving data attribution accuracy and facilitating better model understanding.

Federated data acquisition via experimental design (DAVED) achieves lower prediction error without labeled validation data, optimizing cost-effectively for test-set predictions in decentralized market…

Boost AI model robustness against adversarial attacks by creatively mixing training sample’s frequency amplitude with distractor images, focusing model learning on phase patterns, thus enhancing accur…

DASH combats neural network training’s plasticity loss during warm-starting by selectively forgetting memorized noise while preserving features, improving accuracy and efficiency.

DART-Math tackles LLM limitations in mathematical problem-solving by introducing Difficulty-Aware Rejection Tuning, a novel method that generates high-quality, bias-reduced datasets, resulting in supe…

DARNet: a dual attention network for auditory attention detection surpasses current state-of-the-art models, especially in short decision windows, achieving this with a 91% reduction in parameters.

DarkSAM, a novel prompt-free attack, renders the Segment Anything Model incapable of segmenting objects across diverse images, highlighting its vulnerability to universal adversarial perturbations.

DARG dynamically evaluates LLMs via adaptive reasoning graphs, revealing performance drops with increased complexity and exposing model biases.

DAPE: A novel data-adaptive positional encoding method dynamically adjusts positional information based on input context, improving transformer performance and length generalization.

DAGER: Exact gradient inversion for LLMs; recovers full input text batches precisely.

DA-Ada enhances domain adaptive object detection by using a novel domain-aware adapter that leverages both domain-invariant and domain-specific knowledge for improved accuracy and generalization acros…

D2R2: A novel diffusion-based model for tabular few-shot learning, achieves state-of-the-art results by leveraging semantic knowledge and distance matching.

D-LLM dynamically allocates computing resources during LLM token processing, reducing computational costs and memory usage by up to 50% without sacrificing accuracy.

New D-CPT Law optimizes continual pre-training for LLMs by predicting optimal data mixture ratios, drastically cutting training costs.

CYCLO: A novel cyclic graph transformer excels at multi-object relationship modeling in aerial videos.

CV-VAE: A compatible video VAE enabling efficient, high-quality latent video generation by bridging the gap between image and video latent spaces.