Giovanni Puccetti from ISTI-CNR presented research on linguistic probing of language models like BERT and RoBERTa. The research investigates the ability of these models to encode linguistic properties, linking this ability to outlier parameters. Preliminary work on fine-tuning LLMs in Italian and detecting synthetic news generation was also presented. Why it matters: Understanding the inner workings and linguistic capabilities of LLMs is crucial for improving their reliability and adapting them to diverse languages like Arabic.
The paper introduces Sparse-Quantized Representation (SpQR), a new compression format and quantization technique for large language models (LLMs). SpQR identifies outlier weights and stores them in higher precision while compressing the remaining weights to 3-4 bits. The method achieves less than 1% accuracy loss in perplexity for LLaMA and Falcon LLMs and enables a 33B parameter LLM to run on a single 24GB consumer GPU. Why it matters: This enables near-lossless compression of LLMs, making powerful models accessible on resource-constrained devices and accelerating inference without significant accuracy degradation.
This article discusses methods for handling label noise in deep learning, including extracting confident examples and modeling label noise. Tongliang Liu from the University of Sydney presented these approaches. The talk aimed to provide participants with a basic understanding of learning with noisy labels. Why it matters: As AI models are increasingly trained on large, noisy datasets, techniques for robust learning become crucial for reliable real-world performance.
The paper introduces TimeHUT, a new method for learning time-series representations using hierarchical uniformity-tolerance balancing of contrastive representations. TimeHUT employs a hierarchical setup to learn both instance-wise and temporal information, along with a temperature scheduler to balance uniformity and tolerance. The method was evaluated on UCR, UAE, Yahoo, and KPI datasets, demonstrating superior performance in classification tasks and competitive results in anomaly detection.
Laurent Najman presented the Power Watershed (PW) optimization framework for image and data processing. The PW framework enhances graph-based data processing algorithms like random walker and ratio-cut clustering, leading to faster solutions. It can be adapted for graph-based cost minimization methods and integrated with deep learning networks. Why it matters: This framework could enable more efficient and scalable image and data processing algorithms relevant to computer vision and related fields in the Middle East.
This paper introduces neural Bayes estimators for censored peaks-over-threshold models, enhancing computational efficiency in spatial extremal dependence modeling. The method uses data augmentation to encode censoring information in the neural network input, challenging traditional likelihood-based approaches. The estimators were applied to assess extreme particulate matter concentrations over Saudi Arabia, demonstrating efficacy in high-dimensional models. Why it matters: The research offers a computationally efficient alternative for environmental modeling and risk assessment in the region.