以好奇心为优化目标

| | | | | --- --- | | | α​(x|𝒟t)=βt​I​(s;(x,y)|𝒟t)−𝔼p​(y|x,𝒟t)​[h​(y|𝒟t)],\alpha(x|\mathcal{D}\_{t})=\beta\_{t}I(s;(x,y)|\mathcal{D}\_{t})-\mathbb{E}\_{p(y|x,\mathcal{D}\_{t})}[h(y|\mathcal{D}\_{t})], | | (1) | where s∈𝒮s\in\mathcal{S} denotes the latent parameter(s) of interest, defining the learning space; h​(y|𝒟t)h(y|\mathcal{D}\_{t}) is a problem-dependent potential energy function quantifying the instantaneous regret associated with outcome yy; and βt≥0\beta\_{t}\geq 0 is the curiosity coefficient that sets the exchange rate between information gain and pragmatic cost. [...] β¯T\bar{\beta}\_{T}: the maximum degree of curiosity. Larger curiosity encourages broader exploration and yields more information gain, but also increases the exploration term in the bound. Thus, βt\beta\_{t} controls the trade-off between learning and regret accumulation. ### 6.3 Practical Usability and Implications The cumulative regret bound in Theorem 6.1. ‣ 6 No-Regret Optimization ‣ Cur

Skip to content I’m Dan Oshinsky, and I run Inbox Collective, an email consultancy. I'm here to share what I've learned about doing great work and building amazing teams. Posted on by Dan Oshinsky # Optimize for Curiosity. When you’re building a new strategy, you’ve got choices on what to optimize for: Growth, engagement, or revenue. Or you can make another choice: To optimize solely for curiosity. What might that mean? Recognizing that a good test starts with a great question — so you need to ask as many questions as you can. Finding a team that’s able to step back from the little details to ask, “I know this is a bit different, but what would happen if we tried this?” Being humble enough to avoid doubling down on a good idea that didn’t quite work. [...] Keeping yourself open to new possibilities, even if they don’t seem obvious at first. Understanding that you’ll never have all the answers. The only way forward is to have an open mind and keep asking questions. Whatever

II. CURIOSITY-DRIVEN OPTIMIZATION Curiosity is the drive to actively explore the interesting regions in search space that most improve the model’s pre-dictions or explanations of what is going on in the world. Originally introduced for reinforcement learning , , the curiosity framework has been used for active learning , , to explain certain patterns of human visual attention better than previous approaches , and to explain concepts such as beauty, attention and creativity , . A. Formalizing Interestingness The interestingness of a new observation is the difference between the performance of an adaptive model on the ob-servation history before and after including the new point. [...] V. DISCUSSION The results in section IV demonstrate that curiosity-driven optimization properly handles a number of typical optimiza-tion challenges. Although based on the general and theoret-ically powerful principle of artificial curiosity, our current implementation exhibits certain weaknesses. In partic

Our operationalization of curiosity as information-seeking in response to knowledge gaps (Loewenstein, 1994; Jirout and Klahr, 2012) describes the cognitive processes involved in curiosity and provides direction for identifying potential methods of promoting curiosity in education, including both individual student characteristics and contextual factors. When specifically considering individual factors, being more curious means having a higher preference or threshold for uncertainty, that is, being more likely to explore and seek information in the presence of larger knowledge gaps. Prior research suggests that the relation between uncertainty and curiosity follows an inverted U shape curve, where there is an optimal level of uncertainty that will lead to curiosity and exploration [...] move one to act and direct behavior toward finding specific information (Wentworth and Witryol, 2003). Curiosity can help to develop sustained interests (Hidi and Renninger, 2020), and, in turn, promote

Through her research, Bonawitz explores how curiosity operates like mise en place for learning preparing the mind to absorb, connect, and retain new information. It activates attention, memory, and motivation, setting the stage for deeper understanding. Studies from her lab show that simple practices, like encouraging children to ask more questions, not only increase curiosity but also improve learning outcomes. “Children who are more curious do better in math or reading scores in school. And that's particularly true for students that come from more under-resourced communities or students that might have other challenges associated with school,” Bonawitz says. “So, curiosity is the great equalizer for education.” [...] So that's helping you when you get information figure out how to coordinate it with other memories, with other information you already have, how to feel rewarded from that information, and then helping you encode that information. So, curiosity is the physiological stat

In my mind, curiosity, instinct and the human experience should always come first. Optimise the boring bits, but leave space for exploration, critical thinking, and observing what actually makes people happy. Otherwise, we run the risk of constantly improving but never really imagining. Driving efficiency, but losing the human touch. What’s the point in that? #creativity #optimisation #AI [...] In my mind, curiosity, instinct and the human experience should always come first. Optimise the boring bits, but leave space for exploration, critical thinking, and observing what actually makes people happy. Otherwise, we run the risk of constantly improving but never really imagining. Driving efficiency, but losing the human touch. What’s the point in that? #creativity #optimisation #AI [...] In my mind, curiosity, instinct and the human experience should always come first. Optimise the boring bits, but leave space for exploration, critical thinking, and observing what actually makes people ha