Nixtla’s Post

🚀 Our vision at Nixtla is to become the go-to solution for time series and empower our great community. We also believe that practitioners should easily be able to test and benchmark new ideas. 🙏 This release of NeuralForecast is particularly exciting because it is the first (to my knowledge) open-source implementation of KANS for TimeSeries. Kolmogorov-Arnold Networks (KANs) are an interesting alternative to Multi-Layer Perceptrons (MLPs). While MLPs have fixed activation functions on nodes ("neurons"), KANs have learnable activation functions on edges ("weights").  🤔 Should you start using KANs now for all your forecasting needs? Probably not, but they are worth exploring. 🔥 We are also extremely thankful for the wonderful contributions of the community. A particular shout-out to Jasmine Rienecker and Dr. Fabian Bergermann.

Ever wondered how machine learning is being applied to the field of peripheral nerve interfaces? Well there is a scoping review about that now in TNSRE (https://lnkd.in/eBq57YMN). Couldn't have done it without the help of my co-authors, Mafalda Ribeiro, Leen Jabban, Edith Fang, Karlo Nesovic, Sayeh Bayat, & Benjamin Metcalfe!

Foundational time series model based on patches, evaluated on classification and anomaly detection

Check out this intriguing blog post on "The Surprising Harmfulness of Benign Overfitting for Adversarial Robustness" which delves into the impact of benign overfitting on adversarial vulnerability in machine learning models. The study uncovers unexpected results that show benign overfitting can be detrimental when faced with adversarial manipulation, shedding light on an important aspect of model robustness. Read the full post here: https://bit.ly/3OegL9v [cs.LG].

I'm proud to share our new publication, freely available online open-access: A Benchmarking on Optofluidic Microplastic Pattern Recognition: A Systematic Comparison Between Statistical Detection Models and ML-Based Algorithms Our study undertakes a comparative examination between a statistical detection model and classifiers derived from diverse machine learning paradigms for the purpose of addressing the microplastic detection task. The SDM, a commonly employed method for such problem domains, serves as a benchmark for the evaluation of the effectiveness of machine learning techniques. Additionally, the study probes the performance of machine learning algorithms under conditions deviating from the ideal circumstances in which the model was trained, mirroring real-world scenarios. https://lnkd.in/eGjqPQW6

Artificial Intelligence for Design and Engineering Workflows: In this white paper, learn how artificial intelligence and machine learning can improve design and simulation. http://dlvr.it/TFH7c1

What useful features are exposed by quantum algorithms like Shor’s? And how can we use them to learn from data? Our new paper looks at the engine behind traditional quantum routines and how it can lead to new quantum machine learning heuristics. arxiv.org/abs/2409.00172 Spoiler alert: this is different from your usual QML paper - you won’t find any variational circuits that beat classical ML, nor any proof for quantum advantage.

Delighted to share my paper on "Analyzing The Water Quality Using Machine Learning Techniques," presented at the 23rd ISDA conference by Machine Intelligence Research Labs. #ISDA23 #MachineLearning #WaterQuality

How does quantization impact the performance of LLMs? Only minimal! 🤯 A new study ran 500,000 different evaluations on Meta Llama using different quantization strategies. The impact is <1%, but the benefits are up to 2.4 faster inference and 3.5 model size reduction! 🔥 TL;DR; 💯 Quantized models achieve 99% accuracy recovery compared to full-precision 🚀 Up to 2.4x speedup and 3.5x model size reduction with quantization. 📊 Tested Llama 3.1 8B, 70B, and 405B models on OpenLLM Leaderboard, ArenaHard, HumanEval, and text similarity metrics. 🥇W8A8-FP8 dynamic yields the best results 🤗 Quantized models available on Hugging Face. Blog: https://lnkd.in/d86-AiGG Kudos to Neural Magic for their work on Quantization and comprehensive testing of models! 🤗