Recent benchmarks find In-Context Learning (ICL) outperforms both deep learning and tree-based algorithms on small tabular datasets. However, on larger datasets, ICL for tabular learning suffers in both efficiency and effectiveness. In terms of efficiency, transformers incur linear space and quadratic time complexity w.r.t. context size. In terms of effectiveness, contexts at inference encounter distribution shift compared to contexts from pretraining. We propose MixturePFN, which extends Sparse Mixture of Experts to the state-of-the-art ICL for tabular learning model. Specifically, MixturePFN finetunes a specialized ICL expert on each cluster of tabular data and routes new test samples to appropriate experts at inference. MixturePFN supports constant-size contexts by splitting large training datasets into more manageable clusters. MixturePFN addresses distribution shift by finetuning an expert on each training dataset cluster via bootstrapping. Extensive experimental results shows MixturePFN outperforms 19 baselines both in mean rank and as the Condorcet winner across 36 diverse tabular datasets under both accuracy and F1 score with statistical significance.

Current slideware, such as PowerPoint, reinforces the delivery of linear oral presentations. In settings such as question answering sessions or review lectures, more extemporaneous and dynamic presentations are required. An intelligent system that can automatically identify and display the slides most related to the presenter's speech, allows for more speaker flexibility in sequencing their presentation. We present Quester, a system that enables fast access to relevant presentation content during a question answering session and supports nonlinear presentations led by the speaker. Given the slides contents and notes, the system ranks presentation slides based on semantic closeness to spoken utterances, displays the most related slides, and highlights the corresponding content keywords in slide notes. The design of our system was informed by findings from interviews with expert presenters and analysis of recordings of lectures and conference presentations. In a within-subjects study comparing our dynamic support system with a static slide navigation system during a question answering session, presenters expressed a strong preference for our system and answered the questions more efficiently using our system.

Prior studies of public speaking behavior have demonstrated that public speaking anxiety monotonically decreases with the number of co-presenters giving an oral presentation and increases with the size of the audience. However, speaker "hand off" behavior---the verbal and nonverbal cues used to transition from one speaker to another---and its effect on speaker anxiety and presentation quality has not been systematically studied. In this work we report on two empirical studies of speaker hand-off behavior used during human co-presentations. We find that the cues used for hand-offs during prepared and rehearsed presentations differ significantly from the cues observed in face-to-face conversational turn-taking. We describe two systems that leverage automatic recognition of these verbal and nonverbal cues to drive hand-offs during co-presentations with a life-sized virtual agent.

The fear of forgetting what to say next is a common problem in oral presentation delivery. To maintain good content coverage and reduce anxiety, many presenters use written notes in conjunction with presentation slides. However, excessive use of notes during delivery can lead to disengagement with the audience and low quality presentations. We designed IntelliPrompter, a speech-based note display system that automatically tracks a presenter’s coverage of each slide’s content and dynamically adjusts the note display interface to highlight the most likely next topic to present. We developed two versions of our intelligent teleprompters using Google Glass and computer screen displays. The design of our system was informed by findings from 36 interviews with presenters and analysis of a presentation note corpus. In a within-subjects study comparing our dynamic screen-based and Google Glass note display interfaces with a static note system, presenters and independent judges expressed a strong preference for the dynamic screen-based system.

This paper develops a new rule-based decision support system (RB-DSS) to find the safest solutions for routing, scheduling, and assignment in Hazmat transportation management. To define the safe program in RB-DSS, the accident frequency and severity are estimated for different scenarios of transportation, and they are used to classify the scenarios by a new structure of decision tree (DT), which is proposed to select branching variables at the primary levels according to the experts' perception. The outputs of the DT are stated in the form of if-then rules trained by a multilayer perceptron neural network to generalize the safe programs for Hazmat transportation. To illustrate the performance of this approach, the UK road accident data set is used.