Google's Secrets to Getting People to Adopt A.I.

OVERFIT: AI, Machine Learning, and Deep Learning Made Simple

OVERFIT: AI, Machine Learning, and Deep Learning Made Simple

18:09

HR is no longer just about managing people—it’s about shaping the future of work. Jens Baier, BCG’s HR transformation expert, discusses how AI and shifting employee expectations are forcing companies to rethink talent strategies. From re-recruiting to upskilling employees, HR must adapt to a rapidly changing landscape. Learn More: Jens Baier: https://on.bcg.com/41ca7Gv BCG on People Strategy: https://on.bcg.com/3QtAjro Decoding Global Talent: https://on.bcg.com/4gUC4IT…

рік тому 8:39

MP3•Головна епізоду

This excerpt from "Mental Models," a chapter in the "People + AI Guidebook," focuses on the importance of understanding and managing user mental models when designing AI-powered products. The authors discuss how to set expectations for adaptation, onboard users in stages, plan for co-learning, and account for user expectations of human-like interaction. By carefully considering these factors, product designers can ensure that users form accurate mental models and have a positive experience with AI-powered products.

71 епізодів

Google's Secrets to Getting People to Adopt A.I.

published рік тому

Поширити

MP3•Головна епізоду

71 епізодів

Усі епізоди

1
Does the DIFF Transformer make a Diff? 8:03

16 weeks тому8:03

8:03

Introducing a novel transformer architecture, Differential Transformer, designed to improve the performance of large language models. The key innovation lies in its differential attention mechanism, which calculates attention scores as the difference between two separate softmax attention maps. This subtraction effectively cancels out irrelevant context (attention noise), enabling the model to focus on crucial information. The authors demonstrate that Differential Transformer outperforms traditional transformers in various tasks, including long-context modeling, key information retrieval, and hallucination mitigation. Furthermore, Differential Transformer exhibits greater robustness to order permutations in in-context learning and reduces activation outliers, paving the way for more efficient quantization. These advantages position Differential Transformer as a promising foundation architecture for future large language model development. Read the research here: https://arxiv.org/pdf/2410.05258…

1
Automating Scientific Discovery: ScienceAgentBench 9:49

17 weeks тому9:49

9:49

Introducing, ScienceAgentBench, a new benchmark for evaluating language agents designed to automate scientific discovery. The benchmark comprises 102 tasks extracted from 44 peer-reviewed publications across four disciplines, encompassing essential tasks in a data-driven scientific workflow such as model development, data analysis, and visualization. To ensure scientific authenticity and real-world relevance, the tasks were validated by nine subject matter experts. The paper presents an array of evaluation metrics for assessing program execution, results, and costs, including a rubric-based approach for fine-grained evaluation. Through comprehensive experiments on five LLMs and three frameworks, the study found that the best-performing agent, Claude-3.5-Sonnet with self-debug, could only solve 34.3% of the tasks using expert-provided knowledge. These findings highlight the limitations of current language agents in fully automating scientific discovery, emphasizing the need for more rigorous assessment and future research on improving their capabilities for data processing and utilizing expert knowledge. Read the paper: https://arxiv.org/pdf/2410.05080…

1
Prune This! PyTorch and Efficient AI 8:04

17 weeks тому8:04

8:04

Both sources explain neural network pruning techniques in PyTorch. The first source, "How to Prune Neural Networks with PyTorch," provides a general overview of the pruning concept and its various methods, along with practical examples of how to implement different pruning techniques using PyTorch's built-in functions. The second source, "Pruning Tutorial," focuses on a more in-depth explanation of pruning functionalities within PyTorch, demonstrating how to prune individual modules, apply iterative pruning, serialize pruned models, and even extend PyTorch with custom pruning methods. Read this: https://towardsdatascience.com/how-to-prune-neural-networks-with-pytorch-ebef60316b91 And the PyTorch tutorial: https://pytorch.org/tutorials/intermediate/pruning_tutorial.html…

1
AlexWho? Going Deeper with Deep CNNs 11:50

17 weeks тому11:50

11:50

The source is a chapter from the book "Dive into Deep Learning" that explores the historical development of deep convolutional neural networks (CNNs), focusing on the foundational AlexNet architecture. The authors explain the challenges faced in training CNNs before the advent of AlexNet, including limited computing power, small datasets, and lack of crucial training techniques. They discuss how AlexNet overcame these obstacles by leveraging powerful GPUs, large-scale datasets like ImageNet, and innovative training strategies. The chapter also delves into the architecture of AlexNet, highlighting its similarities to LeNet, and comparing its advantages in terms of depth, activation function, and model complexity control. Finally, the authors emphasize the importance of AlexNet as a crucial step towards the development of the deep networks used today, showcasing its impact on the field of computer vision and deep learning. Read more: https://d2l.ai/chapter_convolutional-modern/alexnet.html…

1
Predicting the Future from the Past: Sequential RNN Stuff 9:47

17 weeks тому9:47

9:47

This text is an excerpt from the "Dive into Deep Learning" book, specifically focusing on the processing of sequential data. The authors introduce the challenges of working with data that occurs in a specific order, like time series or text, and how these sequences cannot be treated as independent observations. They delve into autoregressive models, where future values are predicted based on past values, and highlight the common problem of error accumulation when predicting further into the future. The text discusses the concept of Markov models, where only a limited history is needed to predict future events, as well as the importance of understanding the causal structure of the data. The excerpt then provides a practical example of using linear regression for autoregressive modeling on synthetic time series data and demonstrates the limitations of simple models for long-term prediction. Read more: https://d2l.ai/chapter_recurrent-neural-networks/sequence.html…

1
Google's Secrets to Getting People to Adopt A.I. 8:39

17 weeks тому8:39

8:39

1
LLM Tokenizers, from HFs LNP Course 12:23

18 weeks тому12:23

12:23

This excerpt from Hugging Face's NLP course provides a comprehensive overview of tokenization techniques used in natural language processing. Tokenizers are essential tools for transforming raw text into numerical data that machine learning models can understand. The text explores various tokenization methods, including word-based, character-based, and subword tokenization, highlighting their advantages and disadvantages. It then focuses on the encoding process, where text is first split into tokens and then converted to input IDs. Finally, the text demonstrates how to decode input IDs back into human-readable text. Read more: https://huggingface.co/learn/nlp-course/en/chapter2/4…

1
PyTorch vs Tensorflow: Who Wins in CNN? 11:49

18 weeks тому11:49

11:49

This research paper examines the efficiency of two popular deep learning libraries, TensorFlow and PyTorch, in developing convolutional neural networks. The authors aim to determine if the choice of library impacts the overall performance of the system during training and design. They evaluate both libraries using six criteria: user-friendliness, available documentation, ease of integration, overall training time, overall accuracy, and execution time during evaluation. The paper proposes a novel methodology for comparing these libraries by eliminating external factors that could influence the comparison and focusing solely on the six chosen criteria. The study finds that while both libraries offer similar capabilities, PyTorch is better suited for tasks that prioritize speed and ease of use, while TensorFlow excels in tasks demanding accuracy and flexibility. The authors conclude that the choice of library has a significant impact on both design and performance and that the presented criteria can assist users in selecting the most appropriate library for their specific needs. Read more: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9699128/pdf/sensors-22-08872.pdf…

1
Google's 43 Rules for Machine Learning 9:28

18 weeks тому9:28

9:28

This document provides a comprehensive set of rules for building and deploying machine learning systems, focusing on best practices gleaned from Google’s extensive experience. The document is divided into sections that cover the key stages of the machine learning process, including launching a product without ML, designing and implementing metrics, creating a first pipeline, feature engineering, human analysis, and addressing the challenges of training-serving skew. The rules cover a wide range of topics, from choosing the right objective function to detecting silent failures, and from creating human-understandable features to avoiding feedback loops. The document also offers guidance for navigating the transition from simple to more complex models as a system matures and performance plateaus. Go deeper here: https://developers.google.com/machine-learning/guides/rules-of-ml…

1
Do we Need the Mamba Mindset when LLMs Fail? MoE Mamba and SSMs 11:57

18 weeks тому11:57

11:57

The research paper "MoE-Mamba: Efficient Selective State Space Models with Mixture of Experts" explores a novel approach to language modeling by combining State Space Models (SSMs), which offer linear-time inference and strong performance in long-context tasks, with Mixture of Experts (MoE), a technique that scales model parameters while minimizing computational demands. The authors introduce MoE-Mamba, a model that interleaves Mamba, a recent SSM-based model, with MoE layers, resulting in significant performance gains and training efficiency. They demonstrate that MoE-Mamba outperforms both Mamba and standard Transformer-MoE architectures. The paper also explores different design choices for integrating MoE within Mamba, showcasing promising directions for future research in scaling language models beyond tens of billions of parameters. Read it: https://arxiv.org/abs/2401.04081…

1
Agentic Retrieval Augmented Generation (RAG) systems 7:39

18 weeks тому7:39

7:39

We discuss how to build Agentic Retrieval Augmented Generation (RAG) systems, which use AI agents to retrieve information from various sources to answer user queries. The author details the challenges he faced when building an Agentic RAG system to answer customer support questions, and provides insights into techniques like prompt engineering and structured responses that helped him achieve better results. He also discusses best practices for building effective Agentic RAG systems and touches on the future of AI agents, including multi-agent systems and autonomous agents. Read more: https://arxiv.org/pdf/2408.08435…

1
Let's Get Activated! Why Non-Linear Activation Matters 7:15

18 weeks тому7:15

7:15

Let's get RE(a)L, U! This research paper explores the impact of different activation functions, specifically ReLU and L-ReLU, on the performance of deep learning models. The authors investigate how the choice of activation function, along with factors like the number of parameters and the shape of the model architecture, influence model accuracy across various data domains (continuous, categorical with and without transfer learning). The study concludes that L-ReLU is more effective than ReLU when the number of parameters is relatively small, while ReLU generally performs better with larger models. The paper also highlights the importance of considering the specific data domain and the use of pre-trained models for transfer learning when selecting the most suitable activation function. Read more: https://github.com/christianversloot/machine-learning-articles/blob/main/why-nonlinear-activation-functions-improve-ml-performance-with-tensorflow-example.md…

1
Reviewing Stanford on Linear Regression and Gradient Descent 8:25

19 weeks тому8:25

8:25

This lecture from Stanford University's CS229 course, "Machine Learning," focuses on the theory and practice of linear regression and gradient descent, two fundamental machine learning algorithms. The lecture begins by motivating linear regression as a simple supervised learning algorithm for regression problems where the goal is to predict a continuous output based on a set of input features. The lecture then introduces the cost function used in linear regression, which measures the squared error between the predicted output and the true output. Gradient descent, an iterative algorithm, is then explained as a method to find the parameters that minimize the cost function. Two variants of gradient descent, batch gradient descent and stochastic gradient descent, are discussed with their respective strengths and weaknesses. The lecture concludes with a derivation of the normal equations, an alternative approach to finding the optimal parameters in linear regression that involves solving a system of equations rather than iteratively updating parameters. Watch Andrew Ng teach it at Stanford: https://www.youtube.com/watch?v=4b4MUYve_U8&t=1086s&pp=ygUSdmFuaXNoaW5nIGdyYWRpZW50…

1
Where'd My Gradient Go? It Vanished! 8:39

19 weeks тому8:39

8:39

This video discusses the vanishing gradient problem, a significant challenge in training deep neural networks. The speaker explains how, as a neural network becomes deeper, gradients—measures of how changes in network parameters affect the loss function—can decrease exponentially, leading to a situation where early layers of the network are effectively frozen and unable to learn. This problem arises because common activation functions like the sigmoid function can produce very small derivatives, which compound during backpropagation. The video then explores solutions like using different activation functions (like ReLU) and architectural changes (like residual networks and LSTMs) to mitigate this issue. Watch the video: https://www.youtube.com/watch?v=ncTHBi8a9uA&pp=ygUSdmFuaXNoaW5nIGdyYWRpZW50…

1
Automating Scientific Discovery: ScienceAgentBench 7:38

19 weeks тому7:38