Our goal is to deploy a high-accuracy system starting with zero training examples. We consider an on-the-job setting, where as inputs arrive, we use real-time crowd-sourcing to resolve uncertainty where needed and output our prediction when confident. As the model improves over time, the reliance on crowdsourcing queries decreases. We cast our setting as a stochastic game based on Bayesian decision theory, which allows us to balance latency, cost, and accuracy objectives in a principled way. Computing the optimal policy is intractable, so we develop an approximation based on Monte Carlo Tree Search. We tested our approach on three datasets--named-entity recognition, sentiment classification, and image classification. On the NER task we obtained more than an order of magnitude reduction in cost compared to full human annotation, while boosting performance relative to the expert provided labels.

Our goal is to deploy a high-accuracy system starting with zero training examples. We consider an "on-the-job" setting, where as inputs arrive, we use real-time crowdsourcing to resolve uncertainty where needed and output our prediction when confident. As the model improves over time, the reliance on crowdsourcing queries decreases. We cast our setting as a stochastic game based on Bayesian decision theory, which allows us to balance latency, cost, and accuracy objectives in a principled way. Computing the optimal policy is intractable, so we develop an approximation based on Monte Carlo Tree Search. We tested our approach on three datasets-- named-entity recognition, sentiment classification, and image classification. On the NER task we obtained more than an order of magnitude reduction in cost compared to full human annotation, while boosting performance relative to the expert provided labels. We also achieve a 8% F1 improvement over having a single human label the whole set, and a 28% F1 improvement over online learning.

Our goal is to deploy a high-accuracy system starting with zero training examples. We consider an on-the-job setting, where as inputs arrive, we use real-time crowdsourcing to resolve uncertainty where needed and output our prediction when confident. As the model improves over time, the reliance on crowdsourcing queries decreases. We cast our setting as a stochastic game based on Bayesian decision theory, which allows us to balance latency, cost, and accuracy objectives in a principled way. Computing the optimal policy is intractable, so we develop an approximation based on Monte Carlo Tree Search. We tested our approach on three datasets--named-entity recognition, sentiment classification, and image classification. On the NER task we obtained more than an order of magnitude reduction in cost compared to full human annotation, while boosting performance relative to the expert provided labels.

Are you worried that having hybrid and especially full-time remote employees – even with remote training and virtual mentoring – will undermine junior employee on-the-job learning, integration into company culture, and intra and inter-team collaboration? This issue came up time and time again in my interviews with 47 mid-level and 14 senior leaders at 12 organizations I guided in developing and implementing their strategy for returning to the office and establishing permanent work arrangements for the future of work. If you enjoy video, here's a videocast based on this blog: And if you like audio, here's a podcast based on the blog These leaders acknowledged the reality that the future of work is mainly hybrid, with some staff full-time remote. After all, many high-quality surveys illustrate that 60-70% of all employees want a hybrid schedule permanently after the pandemic. Of the rest, 25-35% want a fully-remote schedule, and only 15-25% want full-time work in the office.

Our goal is to deploy a high-accuracy system starting with zero training examples. We consider an "on-the-job" setting, where as inputs arrive, we use real-time crowdsourcing to resolve uncertainty where needed and output our prediction when confident. As the model improves over time, the reliance on crowdsourcing queries decreases. We cast our setting as a stochastic game based on Bayesian decision theory, which allows us to balance latency, cost, and accuracy objectives in a principled way. Computing the optimal policy is intractable, so we develop an approximation based on Monte Carlo Tree Search.

The rush of intelligent machines and sophisticated analytics into many aspects of work means that trainees are losing opportunities to acquire skills through on-the-job learning (OJL). In medicine, policing, and other fields, people are finding rule-breaking ways to acquire needed expertise out of the limelight. This "shadow learning" is tolerated for the results it produces, but it can exact a personal and an organizational toll. In response, organizations should carefully uncover and study shadow learning; adapt practices that develop organizational, technological, and work designs that enhance OJL; and make intelligent machines part of the solution. It's 6:30 in the morning, and Kristen is wheeling her prostate patient into the OR. Today she's hoping to do some of the procedure's delicate, nerve-sparing dissection herself. The attending physician is by her side, and their four hands are mostly in the patient, with Kristen leading the way under his watchful guidance. The work goes smoothly, the attending backs away, and Kristen closes the patient by 8:15, with a junior resident looking over her shoulder.

Our goal is to deploy a high-accuracy system starting with zero training examples. We consider an “on-the-job” setting, where as inputs arrive, we use real-time crowdsourcing to resolve uncertainty where needed and output our prediction when confident. As the model improves over time, the reliance on crowdsourcing queries decreases. We cast our setting as a stochastic game based on Bayesian decision theory, which allows us to balance latency, cost, and accuracy objectives in a principled way. Computing the optimal policy is intractable, so we develop an approximation based on Monte Carlo Tree Search. We tested our approach on three datasets-- named-entity recognition, sentiment classification, and image classification. On the NER task we obtained more than an order of magnitude reduction in cost compared to full human annotation, while boosting performance relative to the expert provided labels. We also achieve a 8% F1 improvement over having a single human label the whole set, and a 28% F1 improvement over online learning.

Our goal is to deploy a high-accuracy system starting with zero training examples. We consider an "on-the-job" setting, where as inputs arrive, we use real-time crowdsourcing to resolve uncertainty where needed and output our prediction when confident. As the model improves over time, the reliance on crowdsourcing queries decreases. We cast our setting as a stochastic game based on Bayesian decision theory, which allows us to balance latency, cost, and accuracy objectives in a principled way. Computing the optimal policy is intractable, so we develop an approximation based on Monte Carlo Tree Search. We tested our approach on three datasets---named-entity recognition, sentiment classification, and image classification. On the NER task we obtained more than an order of magnitude reduction in cost compared to full human annotation, while boosting performance relative to the expert provided labels. We also achieve a 8% F1 improvement over having a single human label the whole set, and a 28% F1 improvement over online learning.