Time series foundation models (TSFMs) demonstrate impressive zero-shot performance for time series forecasting. However, an important yet underexplored challenge is how to effectively finetune TSFMs on specific downstream tasks. While naive finetuning can yield performance gains, we argue that it falls short of fully leveraging TSFMs' capabilities, often resulting in overfitting and suboptimal performance. Given the diverse temporal patterns across sampling scales and the inherent multi-scale forecasting capabilities of TSFMs, we adopt a causal perspective to analyze finetuning process, through which we highlight the critical importance of explicitly modeling multiple scales and reveal the shortcomings of naive approaches. Focusing on encoder-based TSFMs, we propose MultiScale FineTuning (MSFT), a simple yet general framework that explicitly integrates multi-scale modeling into the finetuning process. Experimental results on three different backbones (MOIRAI, MOMENT and UNITS) demonstrate that TSFMs finetuned with MSFT not only outperform naive and typical parameter efficient finetuning methods but also surpass state-of-the-art deep learning methods. Codes are available at https://github.com/zqiao11/MSFT.

Time series foundation models (TSFMs) demonstrate impressive zero-shot performance for time series forecasting. However, an important yet underexplored challenge is how to effectively finetune TSFMs on specific downstream tasks. While naive finetuning can yield performance gains, we argue that it falls short of fully leveraging TSFMs' capabilities, often resulting in overfitting and suboptimal performance. Given the diverse temporal patterns across sampling scales and the inherent multi-scale forecasting capabilities of TSFMs, we adopt a causal perspective to analyze finetuning process, through which we highlight the critical importance of explicitly modeling multiple scales and reveal the shortcomings of naive approaches. Focusing on encoder-based TSFMs, we propose Multiscale finetuning (MSFT), a simple yet general framework that explicitly integrates multi-scale modeling into the finetuning process. Experimental results on three different backbones (Moirai, Moment and Units) demonstrate that TSFMs finetuned with MSFT not only outperform naive and typical parameter efficient finetuning methods but also surpass state-of-the-art deep learning methods. Codes are available at https://github.com/zqiao11/MSFT.

Time series foundation models (TSFMs) demonstrate impressive zero-shot performance for time series forecasting. However, an important yet underexplored challenge is how to effectively finetune TSFMs on specific downstream tasks. While naive finetuning can yield performance gains, we argue that it falls short of fully leveraging TSFMs' capabilities, often resulting in overfitting and suboptimal performance. Given the diverse temporal patterns across sampling scales and the inherent multi-scale forecasting capabilities of TSFMs, we adopt a causal perspective to analyze finetuning process, through which we highlight the critical importance of explicitly modeling multiple scales and reveal the shortcomings of naive approaches. Focusing on encoder-based TSFMs, we propose Multiscale finetuning (MSFT), a simple yet general framework that explicitly integrates multi-scale modeling into the finetuning process. Experimental results on three different backbones (Moirai, Moment and Units) demonstrate that TSFMs finetuned with MSFT not only outperform naive and typical parameter efficient finetuning methods but also surpass state-of-the-art deep learning methods. Codes are available at https://github.com/zqiao11/MSFT.

Diffusion models have emerged as powerful generative frameworks with widespread applications across machine learning and artificial intelligence systems. While current research has predominantly focused on linear diffusions, these approaches can face significant challenges when modeling a conditional distribution, $P(Y|X=x)$, when $P(X=x)$ is small. In these regions, few samples, if any, are available for training, thus modeling the corresponding conditional density may be difficult. Recognizing this, we show it is possible to adapt the data representation and forward scheme so that the sample complexity of learning a score-based generative model is small in low probability regions of the conditioning space. Drawing inspiration from conditional extreme value theory we characterize this method precisely in the special case in the tail regions of the conditioning variable, $X$. We show how diffusion with a data-driven choice of nonlinear drift term is best suited to model tail events under an appropriate representation of the data. Through empirical validation on two synthetic datasets and a real-world financial dataset, we demonstrate that our tail-adaptive approach significantly outperforms standard diffusion models in accurately capturing response distributions at the extreme tail conditions.

SAN FRANCISCO, April 17 (Reuters) - Billionaire Elon Musk said on Monday he will launch an artificial intelligence (AI) platform that he calls "TruthGPT" to challenge the offerings from Microsoft (MSFT.O) and Google (GOOGL.O). He criticised Microsoft-backed OpenAI, the firm behind chatbot sensation ChatGPT, of "training the AI to lie" and said OpenAI has now become a "closed source", "for-profit" organisation "closely allied with Microsoft". He also accused Larry Page, co-founder of Google, of not taking AI safety seriously. "I'm going to start something which I call'TruthGPT', or a maximum truth-seeking AI that tries to understand the nature of the universe," Musk said in an interview with Fox News Channel's Tucker Carlson aired on Monday. He said TruthGPT "might be the best path to safety" that would be "unlikely to annihilate humans".

We've entered the age of AI. Artificial intelligence is beginning to reshape huge swaths of the global economy, and businesses of all sizes are rushing to deploy the game-changing technology. The AI market will approach a staggering $1.4 trillion by the end of the decade, according to a forecast by Fortune Business Insights. Here are the three companies best positioned to capture sizable portions of this booming industry. Investors' excitement for AI reached a fever pitch after Microsoft (MSFT 0.57%) announced a multibillion-dollar partnership with ChatGPT creator OpenAI in January. Since then, the software giant has moved quickly to add OpenAI's technology to its popular productivity tools and its Bing search engine.

OpenAI and its association with Microsoft Corp. (NASDAQ:MSFT) has perked up interest in artificial intelligence and companies that may have even the remotest ties with the technology. At least one technocrat isn't impressed with what has transpired. What Happened: Tesla CEO Elon Musk lamented how ChatGPT parent OpenAI, a company he co-founded in December 2015, has evolved. He was responding to a tweet by substack journalist and Grit Capital CEO Genevieve Roch-Dector, who noted that Musk views AI as "one of the biggest risks" to civilization and needs to be regulated. "He co-founded AI," she pointed out.

In this article, I would like to start with a recent announcement that Microsoft (NASDAQ:MSFT) made and then show how, even though it may be of little relevance, it offers once again the opportunity to understand how Microsoft runs its business and, most important, defends its wide moat. I really enjoy carrying out this kind of research, especially when I have to deal with a very large company such as Microsoft. In fact, I think that very often, understanding well how one particular choice works, enables me to get a grasp of the whole company better than if I were to analyze only its financials without diving into some of its operations. Let's get to the announcement: Microsoft is launching Microsoft Designer, a graphic design app in Microsoft 365 that helps users create social media posts, invitations, digital postcards, graphics, and more, all in a flash. The most important feature is that Microsoft Designer is powered by AI technology, including DALL E 2 by OpenAI, which enables to instantly generate a variety of designs with minimal effort.

Given the segment's solid growth prospects, we think it could be wise to bet on quantum computing stocks Microsoft (MSFT), Alphabet (GOOGL), International Business Machines (IBM), and Hitachi (HTHIY). Given quantum computing's growth prospects, tech giants are investing significantly in the space to grab market share. Technology giant MSFT's broad product portfolio includes personal computers, tablets, gaming and entertainment consoles, and related accessories. On Jan. 25, 2022, Satya Nadella, chairman and CEO, MSFT, said, "As tech as a percentage of global GDP continues to increase, we are innovating and investing across diverse and growing markets, with a common underlying technology stack and an operating model that reinforces a common strategy, culture, and a sense of purpose." Mountain View, Calif.-based GOOGL provides online advertising services in the United States, Europe, the Middle East, Africa, Asia-Pacific, Canada, and Latin America.

In a recent article published in CoinDesk, I outlined some of the key challenges of quant strategies for crypto assets. Creating predictive models and quant strategies for crypto assets is a fascinating challenges and one that present very novel difficulties compared to traditional capital markets. As we have been building more machine learning(ML)-based predictive models at IntoTheBlock, we have encountered several hurtles that fall outside traditional machine learning and quant methodologies. One of those challenges is what I referred to in the article as the "retraining dilemma". ML-based predictive models for financial assets such as cryptocurrencies are fundamentally based in supervised learning methods.