An official website of the United States government
ABSTRACT This paper studies the transfer learning problem for convolutional neural network models. A phase transition phenomenon has been empirically validated: The convolutional layer shifts from general to specific with respect to the target task as its depth increases. The paper suggests measuring the generality of convolutional layers through an easy‐to‐compute and tuning‐free statistic named projection correlation. The non‐asymptotic upper bounds for the estimation error of the proposed generality measure have been provided. Based on this generality measure, the paper proposes a forward‐adding‐layer‐selection algorithm to select general layers. The algorithm aims to find a cut‐off in the pre‐trained model according to where the phase transition from general to specific happens. Then, we propose to transfer only the general layers as specific layers can cause overfitting issues and hence hurt the prediction performance. The proposed algorithm is computationally efficient and can consistently estimate the true beginning of phase transition under mild conditions. Its superior empirical performance has been justified by various numerical experiments.
more »
« less
This content will become publicly available on October 6, 2026
Considerations in Pursuing Reaction Scope Generality
Abstract The term “generality” has recently been popularized in synthetic chemistry, owing largely to the increasing use of high‐throughput technology for producing vast quantities of data and the emergence of data science tools to plan and interpret these experiments. Despite this, the term has not been clearly defined, and there is no standardized approach toward developing a method with a diverse (general) scope. This minireview will examine different emerging strategies toward achieving generality using selected examples and aims to give the reader an overview of modern workflows that have been used to expedite this pursuit.
more »
« less
- Award ID(s):
- 2202693
- PAR ID:
- 10645101
- Publisher / Repository:
- Wiley
- Date Published:
- Journal Name:
- Angewandte Chemie International Edition
- Volume:
- 64
- Issue:
- 41
- ISSN:
- 1433-7851
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Critical data studies, a body of emergent research that draws on surveillance studies and other fields, investigates datafication, the increasing mediation of many forms of sociality by data-intensive, networked computation. Such research draws on well-trodden criticisms of the representational capacities of data and has recently offered the term “data justice” to direct this scholarly formation toward the harms of datafication. By failing to explicitly foreground the way that the capture and consultation of data constitutes a tactic by which the state sorts, controls, and limits the freedom of minoritized peoples, critical data studies substitutes an interest in describing forms of injustice for a commitment toward its undoing. I use McKittrick’s (2021) term “seeking liberation” to orient both surveillance studies and critical data studies scholarship away from mere description of the practices of data-intensive computation and surveillance and toward a shared project of justice for minoritized peoples.more » « less
-
Abstract The relationship between people, place, and data presents challenges and opportunities for science and society. While there has been general enthusiasm for and work toward Findable, Accessible, Interoperable, and Reusable (FAIR) data for open science, only more recently have these data-centric principles been extended into dimensions important to people and place—notably, the CARE Principles for Indigenous Data Governance, which affect collective benefit, authority to control, responsibility, and ethics. The FAIR Island project seeks to translate these ideals into practice, leveraging the institutional infrastructure provided by scientific field stations. Starting with field stations in French Polynesia as key use cases that are exceptionally well connected to international research networks, FAIR Island builds interoperability between different components of critical research infrastructure, helping connect these to societal benefit areas. The goal is not only to increase reuse of scientific data and the awareness of work happening at the field stations but more generally to accelerate place-based research for sustainable development. FAIR Island works reflexively, aiming to scale horizontally through networks of field stations and to serve as a model for other sites of intensive long-term scientific study.more » « less
-
Given the fundamental role of working memory (WM) in all domains of cognition, a central question has been whether WM is domain-general. However, the term ‘domain-general’ has been used in different, and sometimes misleading, ways. By reviewing recent evidence and biologically plausible models of WM, we show that the level of domain-generality varies substantially between three facets of WM: in terms of computations, WM is largely domain-general. In terms of neural correlates, it contains both domain-general and domain-specific elements. Finally, in terms of application, it is mostly domain-specific. This variance encourages a shift of focus towards uncovering domain-general computational principles and away from domain-general approaches to the analysis of individual differences and WM training, favoring newer perspectives, such as training-as-skill-learning.more » « less
-
Abstract Organisms gain by anticipating future changes in the environment. Those environmental changes often follow stochastic trends. The steeper the slope of the trend, the more likely the trend’s momentum carries the future trend in the same direction. This article presents a simple biological circuit that measures the momentum, providing a prediction about future trend. The circuit calculates the momentum by the difference between a short-term and a long-term exponential moving average. The time lengths of the two moving averages can be adjusted by changing the decay rates of state variables. Different time lengths for those averages trade off between errors caused by noise and errors caused by lags in predicting a change in the direction of the trend. Prior studies have emphasized circuits that make similar calculations about trends. However, those prior studies embedded their analyses in the details of particular applications, obscuring the simple generality and wide applicability of the approach. The model here contributes to the topic by clarifying the great simplicity and generality of anticipation for stochastic trends. This article also notes that, in financial analysis, the difference between moving averages is widely used to predict future trends in asset prices. The financial measure is called the moving average convergence–divergence indicator. Connecting the biological problem to financial analysis opens the way for future studies in biology to exploit the variety of highly developed trend models in finance.more » « less
