Uncertainty estimation is critical in high-stakes machine learning
applications. One effective way to estimate uncertainty is conformal
prediction, which can provide predictive inference with statistical coverage
guarantees. We present a new conformal regression method, Spline Prediction
Intervals via Conformal Estimation (SPICE), that estimates the conditional
density using neural-network-parameterized splines. We prove universal
approximation and optimality results for SPICE, which are empirically validated
by our experiments. SPICE is compatible with two different efficient-to-compute
conformal scores, one oracle-optimal for marginal coverage (SPICE-ND) and the
other asymptotically optimal for conditional coverage (SPICE-HPD). Results on
benchmark datasets demonstrate SPICE-ND models achieve the smallest average
prediction set sizes, including average size reductions of nearly 50% for some
datasets compared to the next best baseline. SPICE-HPD models achieve the best
conditional coverage compared to baselines. The SPICE implementation is made
available.

通过神经网络参数化样条估计条件密度的 SPICE 方法是一种新的与可靠统计覆盖保证相容的一致性回归方法，具有普遍逼近和最优性结果，适用于两种高效计算的一致性评分，能够提供较小的平均预测区间并保证最佳条件覆盖。

优化和高效预测集的合规化深度样条

Conformalized Deep Splines for Optimal and Efficient Prediction Sets

There is considerable interest in the task of automatically generating image
captions. However, evaluation is challenging. Existing automatic evaluation
metrics are primarily sensitive to n-gram overlap, which is neither necessary
nor sufficient for the task of simulating human judgment. We hypothesize that
semantic propositional content is an important component of human caption
evaluation, and propose a new automated caption evaluation metric defined over
scene graphs coined SPICE. Extensive evaluations across a range of models and
datasets indicate that SPICE captures human judgments over model-generated
captions better than other automatic metrics (e.g., system-level correlation of
0.88 with human judgments on the MS COCO dataset, versus 0.43 for CIDEr and
0.53 for METEOR). Furthermore, SPICE can answer questions such as `which
caption-generator best understands colors?' and `can caption-generators count?'

提出了基于场景图谱的自动生成图像标题评估指标 SPICE，其捕捉人类判断模型生成标题的能力优于现有指标，并能回答哪个标题生成器更懂颜色和能否计数等问题。

SPICE：语义命题图像字幕评估

SPICE: Semantic Propositional Image Caption Evaluation

Any national cuisine is a sum total of its variety of regional cuisines,
which are the cultural and historical identifiers of their respective regions.
India is home to a number of regional cuisines that showcase its culinary
diversity. Here, we study recipes from eight different regional cuisines of
India spanning various geographies and climates. We investigate the phenomenon
of food pairing which examines compatibility of two ingredients in a recipe in
terms of their shared flavor compounds. Food pairing was enumerated at the
level of cuisine, recipes as well as ingredient pairs by quantifying flavor
sharing between pairs of ingredients. Our results indicate that each regional
cuisine follows negative food pairing pattern; more the extent of flavor
sharing between two ingredients, lesser their co-occurrence in that cuisine. We
find that frequency of ingredient usage is central in rendering the
characteristic food pairing in each of these cuisines. Spice and dairy emerged
as the most significant ingredient classes responsible for the biased pattern
of food pairing. Interestingly while individual spices contribute to negative
food pairing, dairy products on the other hand tend to deviate food pairing
towards positive side. Our data analytical study highlighting statistical
properties of the regional cuisines, brings out their culinary fingerprints
that could be used to design algorithms for generating novel recipes and recipe
recommender systems. It forms a basis for exploring possible causal connection
between diet and health as well as prospection of therapeutic molecules from
food ingredients. Our study also provides insights as to how big data can
change the way we look at food.

通过对印度八大不同地域、气候的菜系的食谱进行研究，本研究发现了每个菜系都遵循着负的食物配对模式，表明共享风味化合物的两种材料在一个菜系中摆放位置越远；同时揭示了香料和乳制品是负的食物配对模式发生的主要材料类别，而乳制品可能会使食物配对在常规基础上向着积极的方向偏移。