Tabular data (or tables) are the most widely used data format in machine learning (ML). However, ML models often assume the table structure keeps fixed in training and testing. Before ML modeling, heavy data cleaning is required to merge disparate tables with different columns.

Technical quality: The math seems correct, though in general showing a few more intermediate steps in derivations would make the work of the reader easier (can push the proof to the supplement to make room). In the proof of Proposition 1 for instance, it would be nice to have more detail on how the final equality is obtained. The experiments compare to an SVM baseline, but not to the parameter learning done in [9]. Specifically, [9] does parameter learning for the spin glass model and shows that it achieves an error of 1.8% compared to an SVM's 8.2%. What is different about the learning done in this work (besides the use of a different probabilistic model)?

Summary: Authors propose a novel multi-branch network with a loss function that uses distillation from a combined branch to distill into individual branches. The technique is motivated by the idea that Teacher-Student knowledge distillation is a two-step process often requiring a large pre-trained teacher. Their method builds a teacher, out of weighted ensemble and uses that to train the network. They are able to show that the combined network (ONE-E) is far superior to standalone networks, and the individual branch (ONE) is also better than its counterpart (i.e if it were trained without any of the loss functions and the branches). Pros: 1. Excellent write-up This is a very well written paper.

Summary: This paper develops a model that can capture player-interactions from group comparisons (team-play win/loss info). In an effort to address higher-order interactions with a reasonable size of data set, it then proposes a latent factor model and the sample complexity analysis for the model is done under certain scenarios. Experiments are conducted on real-world on-line game datasets, comparing the win/loss prediction accuracy of the proposed approach to the prior methods such as BTL [12] and Trueskill [11]. Detailed comments: The paper studies an interesting problem, and investigates the role of player-interactions which has been out of reach in the literature. One noticeable observation found in the paper is that the proposed approach may be able to identify the best team members with good chemistry, as suggested in Table 3.

Agents that "cooperate" with each other can take on many forms. The algorithmic architecture described herein addresses one of these forms, that of a team of computer agents that share many aspects of the same "brain.".

Summary The paper proposes a collaborative learning strategy in which multiple classifier heads of the same deep network are trained simultaneously on the same data. Unlike ensembling, knowledge distillation, multi-task learning or training auxiliary classifiers, this seems more effective from the point of view of training time and memory utilization. In addition, the proposed training objective improves generalization and makes the network more robust to label noise. For the intermediate representations that are shared, the authors propose a backpropagation-rescaling technique so as to control the variance of gradients being propagated to the previous layers. Empirical results demonstrate the efficacy of the method w.r.t.

Existing datasets for tabular question answering typically focus exclusively on text within cells. However, real-world data is inherently multimodal, often blending images such as symbols, faces, icons, patterns, and charts with textual content in tables. With the evolution of AI models capable of multimodal reasoning, it is pertinent to assess their efficacy in handling such structured data. This study investigates whether current AI models can perform knowledge-aware reasoning on multimodal structured data. We explore their ability to reason on tables that integrate both images and text, introducing MMTabQA, a new dataset designed for this purpose. Our experiments highlight substantial challenges for current AI models in effectively integrating and interpreting multiple text and image inputs, understanding visual context, and comparing visual content across images. These findings establish our dataset as a robust benchmark for advancing AI's comprehension and capabilities in analyzing multimodal structured data.

We present region-based, fully convolutional networks for accurate and efficient object detection. In contrast to previous region-based detectors such as Fast/Faster R-CNN [7, 19] that apply a costly per-region subnetwork hundreds of times, our region-based detector is fully convolutional with almost all computation shared on the entire image. To achieve this goal, we propose position-sensitive score maps to address a dilemma between translation-invariance in image classification and translation-variance in object detection. Our method can thus naturally adopt fully convolutional image classifier backbones, such as the latest Residual Networks (ResNets) [10], for object detection. We show competitive results on the PASCAL VOC datasets (e.g., 83.6% mAP on the 2007 set) with the 101-layer ResNet. Meanwhile, our result is achieved at a test-time speed of 170ms per image, 2.5-20 faster than the Faster R-CNN counterpart.

This is the post about the work I've done as a research assistant at Stanford. I was lucky to have Tapan Mukerji as my advisor, his guidance helped me to avoid a lot of pitfalls and keep going when I felt stuck. This post is based on my paper. The task I was working on was seismic facies classification with deep learning, and I'd like to start with a high-level overview of why this is a problem worth solving. Seismic data is used to understand the subsurface structure and, ideally, to quantify some properties indicative of where hydrocarbons may be deposited. Figure 1 shows what seismic data may look like.