Introducing Chronos-2: From univariate to universal time series forecasting

Good news! Impressive!

How much will e.g. economics benefit from foundation models? I remember having studied econometrics a few decades ago.

"Chronos-2 can forecast single time series, multiple related time series, and time series influenced by external factors, all without needing extra training. The model uses in-context learning and a group attention feature to understand how different time series relate to each other and to factor in outside influences like weather or sales promotions.

Amazon trained Chronos-2 on synthetic data since real-world datasets with complex relationships between variables are hard to find.

Chronos-2 beat existing forecasting models by wide margins on two major benchmarks, winning over 90 percent of head-to-head comparisons against its predecessor, Chronos-Bolt. The model’s weights are now openly available, and earlier versions have been downloaded over 600 million times from Hugging Face." (Data Points newsletter)

"... time series foundation models (TSFMs) ...

Despite their success, existing TSFMs have a key limitation: they support only univariate forecasting, predicting a single time series at a time. Although univariate forecasting is important, many scenarios require additional capabilities.

Real-world forecasting problems often involve predicting multiple coevolving time series simultaneously (multivariate forecasting) or incorporating external factors that influence outcomes (covariate-informed forecasting). For example, cloud infrastructure metrics such as CPU usage, memory consumption, and storage I/O evolve together and benefit from joint modeling. Likewise, retail demand is heavily influenced by promotional activities, while energy consumption patterns are driven by weather conditions. ..."

Introducing Chronos-2: From univariate to universal forecasting - Amazon Science "In-context learning enables a model that can solve forecasting tasks with an arbitrary number of dimensions in a zero-shot manner."

High-level depiction of Chronos. Left: Input time series is scaled and quantized to obtain a sequence of tokens. Center: The tokens are fed into a language model, which is trained using the cross-entropy loss. Right: During inference, tokens are sampled autoregressively from the model and mapped back to numerical values. (Source)

Asta DataVoyager: Data-driven discovery and analysis of structured experimental data for scientists

Good news! How much will this accelerate and boost scientific research?

".. Experimental logs live in spreadsheets; instrument readings arrive as CSVs; and results tables pile up across projects. Turning those structured files into answers takes time and often requires advanced programming skills to be done efficiently.

To fill the gap, we’re launching DataVoyager in Asta, our ecosystem for scientific research agents. Built to address the challenges scientists face in drilling down into structured datasets, Asta DataVoyager delivers data-driven discovery and analysis capabilities that allow you to ask questions about structured files in plain language and get clearly cited, explainable answers with copyable code, clear visuals, and a concise, well-supported summary. ...

Users upload a dataset and ask a question (e.g., “Which treatment shows the most improvement after week 6?”), along with an optional prompt to establish context so that Asta DataVoyager makes better initial choices. ...

Moreover, Asta DataVoyager allows teams to stay in full control of their data—they can delete datasets at any time from Asta’s hosted portal or secure on-premises, datacenter, and private cloud deployments. ..."

"... Asta DataVoyager is a trusted AI collaborator—one that lets researchers make queries about data in natural language and get transparent, reproducible answers they can act on. It was designed from the start to be intuitive for users, regardless of their comfort level working with dataset analysis tooling. ...

Users upload a dataset in CSV, Excel (.xlsx), JSON (.json/.jsonl), HDF5, TSV, or Parquet format and ask a question (e.g., “Which treatment arm shows the steepest improvement after week 6?”), along with an optional prompt to establish context (e.g., “use these units, measurement cadence, treatment conditions, and outcome variables”) so that Asta DataVoyager makes better initial choices.

Asta DataVoyager then outputs: 

• A crisp answer to the user’s question, written for scientists
• Copyable visuals that make the finding understandable at a glance
• Copyable code that reproduces the analysis
• A methods section that documents assumptions, detailed reasoning steps, and statistical tests conducted—so users can cite the procedure or adapt it

..."

Asta DataVoyager, fluid benchmarking, and build your own AskOlmo bot

Asta DataVoyager: Data-driven discovery and analysis

Scientists Find Optimal Balance of Data Storage and Time for hash tables after 70 years

Recommendable!

After trial and error, they discovered and confirmed that the theoretical lower bound was the same as the previously established upper bound. Case closed! 😊

"... Hash tables are among the oldest, simplest, fastest and most widely used data structures today. They’re designed to perform three basic operations: insertions, which add new items to the database; queries, which access an item or check to see whether it exists; and deletions. ..."

From the abstract:

"For nearly six decades, the central open question in the study of hash tables has been to determine the optimal achievable tradeoff curve between time and space. State-of-the-art hash tables offer the following guarantee: If keys/values are Theta(log n) bits each, then it is possible to achieve constant-time insertions/deletions/queries while wasting only O(loglog n) bits of space per key when compared to the information-theoretic optimum. Even prior to this bound being achieved, the target of O(loglog n) wasted bits per key was known to be a natural end goal, and was proven to be optimal for a number of closely related problems (e.g., stable hashing, dynamic retrieval, and dynamically-resized filters).

This paper shows that O(loglog n) wasted bits per key is not the end of the line for hashing. In fact, for any k \in [log* n], it is possible to achieve O(k)-time insertions/deletions, O(1)-time queries, and O(\log^{(k)} n) wasted bits per key (all with high probability in n). This means that, each time we increase insertion/deletion time by an \emph{additive constant}, we reduce the wasted bits per key \emph{exponentially}. We further show that this tradeoff curve is the best achievable by any of a large class of hash tables, including any hash table designed using the current framework for making constant-time hash tables succinct."

Scientists Find Optimal Balance of Data Storage and Time | Quanta Magazine Seventy years after the invention of a data structure called a hash table, theoreticians have found the most efficient possible configuration for it.

On the Optimal Time/Space Tradeoff for Hash Tables (open access)

Tight Cell-Probe Lower Bounds for Dynamic Succinct Dictionaries (open access)

What is wrong with this chart?

Data science is hard!

It does not show how the deaths relate to the dramatically increased world population over this 100 year period. 

Either a right hand Y-axis showing the world population should have been added or the death should have been expressed as per population (e.g. per 100,000 population). Then, the chart would have been a lot more impressive, I bet.

On Will we run out of data? An analysis of the limits of scaling datasets in Machine Learning. Scarcely!

I have not read this study, but I have some hunches! None of the authors is familiar to me. They hail from the University of Aberdeen (Scotland) and U of Tübingen (Germany). I might be very wrong, but both universities do not seem to be terribly known for main and relevant research related to  machine learning. The senior author, i.e. Anson Ho, has a total lifetime citation count of 25.

1. There is probably some simplistic trend extrapolation involved
2. The authors invoke a distinction between "high-quality language data" and other data. Well, such a distinction is usually riddled with ambiguity!
3. Who said that ever larger models need progressively larger datasets as well? Perhaps, better future algorithms make the need for larger datasets and larger models less relevant
4. Human ingenuity can handle this not least with synthetic data that can be produced in any amount and quality

[2211.04325] Will we run out of data? An analysis of the limits of scaling datasets in Machine Learning

Credits: The Batch by Andrew Ng

More synthetic data for advancing AI and machine learning

Probably, we will see much more synthetic data generated in the future. It's cheaper and more abundant than real data. Can be customized in many ways. Avoids or alleviate hot button issues like gender or skin color bias. 

Similar approaches have been around for at least the past two decades, e.g. data augmentation or toy examples/datasets. 

Caltech: How to Make Sense of Data

Recommendable! Find out more about breath biomarkers, black holes, gravitational waves, embedded glucose sensors for diabetic patients ...

How to Make Sense of Data — Caltech Magazine Caltech scientists and engineers are designing the sensors that gather data and the algorithms that translate data to communicate and work together in a process they call Sensing to Intelligence