# learning_from_rules_generalizing_labeled_exemplars__b52a19b7.pdf

Published as a conference paper at ICLR 2020

LEARNING FROM RULES GENERALIZING LABELED EXEMPLARS

Abhijeet Awasthi Sabyasachi Ghosh Rasna Goyal Sunita Sarawagi Department of Computer Science and Engineering Indian Instiute of Technology Bombay Mumbai, Maharashtra 400076, India {awasthi,sghosh,goyalrasna,sunita}@cse.iitb.ac.in

In many applications labeled data is not readily available, and needs to be collected via pain-staking human supervision. We propose a rule-exemplar method for collecting human supervision to combine the efficiency of rules with the quality of instance labels. The supervision is coupled such that it is both natural for humans and synergistic for learning. We propose a training algorithm that jointly denoises rules via latent coverage variables, and trains the model through a soft implication loss over the coverage and label variables. The denoised rules and trained model are used jointly for inference. Empirical evaluation on five different tasks shows that (1) our algorithm is more accurate than several existing methods of learning from a mix of clean and noisy supervision, and (2) the coupled rule-exemplar supervision is effective in denoising rules.

1 INTRODUCTION

With the ever-increasing reach of machine learning, a common hurdle to new adoptions is the lack of labeled data and the pain-staking process involved in collecting human supervision. Over the years, several strategies have evolved. On the one hand are methods like active learning and crowdconsensus learning that seek to reduce the cost of supervision in the form of per-instance labels. On the other hand is the rich history of rule-based methods (Appelt et al., 1993; Cunningham, 2002) where humans code-up their supervision as labeling rules. There is growing interest in learning from such efficient, albiet noisy, supervision (Ratner et al., 2016; Pal & Balasubramanian, 2018; Bach et al., 2019; Sun et al., 2018; Kang et al., 2018). However, clean task-specific instance labels continue to be critical for reliable results (Goh et al., 2018; Bach et al., 2019) in spite of easy availability of pre-trained models (Sun et al., 2017; Devlin et al., 2018).

In this paper we propose a unique blend of cheap coarse-grained supervision in the form of rules and expensive fine-grained supervision in the form of labeled instances. Instead of supervising rules and instance labels independently, we propose that each labeling rule be attached with exemplars of where the rule correctly fires . Thus, the rule can be treated as a noisy generalization of those exemplars. Often rules are coded up only after inspecting data. As a human inspects instances, he labels them, and then generalizes them to rules. Thus, humans provide paired supervision of rules and exemplars demonstrating correct deployment of that rule. We explain further with two illustrative applications. Our examples below are from the text domain because rules have been traditionally used in many NLP tasks, but our learning algorithm is agnostic to how rules are expressed.

Sentiment Classification Consider an instance I highly recommend this modest priced cellular phone that a human inspects for a sentiment labeling task. After labeling it as positive, he can easily generalize it to a rule Contains highly recommend positive label. This rule generalizes to several more instances, thereby eliminating the need of per-instance labeling on those. However, the label assigned by this rule on unseen instances may not be as reliable as the explicit label on this specific exemplar it generalized. For example, it misfires on I would highly recommend this phone if it weren t for their poor service.

Code and datasets available at https://github.com/awasthiabhijeet/Learning-From-Rules

Published as a conference paper at ICLR 2020

Slot-filling Consider a slot-filling task on restaurant reviews over labels like cuisine, location, and time. When an annotator sees an instance like: what chinese restaurants in this city have good reviews?, after labeling token chinese as cuisine, he generalizes it to a rule: (.*ese|.*ian|mexican) restaurants (cuisine) restaurants. This rule matches hundreds of instances in the unlabeled set, but could wrongly label a phrase like these restaurants. Our focus in this paper is developing algorithms for training models under such coupled rule-exemplar supervision. Our main challenge is that the labels induced by the rules are more noisy than instance-level supervised labels because humans tend to over generalize (Tessler & Goodman, 2019) as we saw in the illustrations above. Learning with noisy labels with or without additional clean data has been a problem of long-standing interest in ML (Khetan et al., 2018; Zhang & Sabuncu, 2018; Ren et al., 2018b; Veit et al., 2017; Shen & Sanghavi, 2019). However, we seek to design algorithms that better capture rule-specific noise with the help of exemplars around which we have supervision that the rule fired correctly. We associate a latent random variable on whether a rule correctly covers an instance, and jointly learn the distribution among the label and all cover variables. This way we simultaneously train the classifier with corrected rule-label examples, and restrict over-generalized rules. The denoised rules are used during inference to further boost accuracy of the trained model. In summary our contributions in this paper are as follows:

Our contributions (1) We propose the paradigm of supervision in the form of rules generalizing labeled exemplars that is natural in several applications. (2) We design a training method that simultaneously denoises over-generalized rules via latent coverage variables, and trains a classification model with a soft implication loss that we introduce. (3) Through experiments on five tasks spanning question classification, spam detection, sequence labeling, and record classification we show that our proposed paradigm of supervision enables an effective synergy between rule-level and instance-level supervision. (4) We compare our algorithm to several recent frameworks for learning with noisy supervision and constraints, and show much better results with our method.

2 TRAINING WITH RULES AND EXEMPLARS

We first formally describe the problem of learning from rules generalizing examplars on a classification task. Let X denote the space of instances and Y = {1, . . . , K} denote the space of class labels. Let the set of labeled examples be L = {(x1, ℓ1, e1), . . . , (xn, ℓn, en)} where xi X is an instance, ℓi Y is its user-provided label, and ei {R1, . . . , Rm, } denotes that xi is an exemplar for rule ei. Some labeled instances may not be generalized to rules and for them ei = . Also, a rule can have more than one exemplar associated with it. Each rule Rj could be a blackbox function Rj : x 7 {ℓj, } that takes as input an instance x X and assigns it either label ℓj or no-label. When the ith labeled instance is an exemplar for rule Rj (that is, ei = Rj), the label of the instance ℓi should be ℓj. Additionally, we have a different set of unlabeled instances U = {xn+1, . . . , x N}. The cover set Hj of rule Rj is the set of all instances in U L for which Rj assigns a noisy label ℓj. An instance may be covered by more than one rule or no rule at all, and the labels provided by these rules may be conflicting. Our goal is to train a classification model Pθ(y|x) using L and U to maximize accuracy on unseen test instances. A baseline solution is to use Rj to noisily label the covered U instances using majority or other consensus method of resolving conflicts. We then train Pθ(y|x) on the noisy labels using existing algorithms for learning from noisy and clean labels (Veit et al., 2017; Ren et al., 2018b). However, we expect to be able to do better by learning the systematic pattern of noise in rules along with the classifier Pθ(y|x).

Our noise model on Rj A basic premise of our learning paradigm is that the noise induced by a rule Rj is due to over-generalizing the exemplar(s) seen when creating the rule. And, there exists a smaller neighborhood closer to the exemplar(s) where the noise is zero. We model this phenomenon by associating a latent Bernoulli random variable rji for each instance xi in the stated cover set Hj of each rule Rj. When rji = 1, rule Rj has not over-generalized on xi, and there is no noise in the label ℓj that Rj assigns to xi. When rji = 0 we flag an over-generalization, and abstain from labeling xi as ℓj suspecting it to be too noisy. We call rjis as the latent coverage variables. We propose to learn the distribution of rj using another network with parameters φ that outputs the probability Pjφ(rj|x) that rj = 1. We then seek to jointly learn Pθ(y|x) and Pjφ(rj|x) to model the distribution over the true label y and true coverage rj for each rule j and each x in Hj. Thus

Published as a conference paper at ICLR 2020

Pjφ plays the role of restricting a rule Rj so that rj is not necessarily 1 for all instances in its cover set Hj

Figure 1: Restricting over-generalized rules

An example We make our discussion concrete with an example. Figure 1 shows a two-dimensional X space with labeled points L denoted as red crosses and blue circles, unlabeled points as dots, and the true labels as background color of the region. We show two rule-exemplar pairs: (x1, y1 = red, R1), (x2, y2 = blue, R2) with bold boundaries. Clearly, both rules R1, R2 have over-generalized to the wrong region. If we train a classifier with many examples in H1 H2 wrongly labeled by rules, then even with a noise tolerant loss function like Zhang & Sabuncu (2018), the classifier Pθ(y|x) might be misled. In contrast, what we hope to achieve is to learn the Pjφ(rj|x) distribution using the limited labeled data and the overlap among the rules such that Pr(rj|x) predicts a value of 0 for examples wrongly covered. Such examples are then excluded from training Pθ. The dashed boundaries indicate the revised boundaries of Rjs that we can hope to learn based on consensus on the labeled data and the set of rules. Even after such restriction, Rjs are useful for training the classifier because of the unlabeled points inside the dashed regions that get added to the labeled set.

2.1 HOW WE JOINTLY LEARN Pθ AND Pjφ In general we will be provided with several rules with arbitrary overlap in the set of labeled L and unlabeled examples U that they cover. Intuitively, we want the label distribution Pθ(y|x) to correctly restrict the coverage distribution Pjφ(rj|x), which in turn can provide clean labels to instances in U that can be used to train Pθ(y|x). We have two types of supervision in our setting. First, individually for each of Pθ(y|x) and Pjφ(rj|x) we have ground truth values of y and rj for some instances. For the Pθ(y|x) distribution, supervision on y is provided by the human labeled data L, and we use these to define the usual log-likelihood as one term in our training objective:

max θ LL(θ) = max θ

(xi,ℓi) L log Pθ(ℓi|xi) (1)

For learning the distribution Pjφ(rj|x) over the coverage variables, the only sure-shot labeled data is that rji = 1 for any xi that is an exemplar of rule Rj and rji = 0 for any xi Hj whose label ℓi is different from ℓj. For other labeled instances xi covered with rules Rj with agreeing labels, that is ℓi = ℓj we do not strictly require that rji = 1. In the example above the corrected dashed red boundary excludes a red labeled point to reduce its noise on other points. However, if the number of labeled exemplars are too few, we regularize the networks towards more rule firings, by adding a noise tolerant rji = 1 loss on the instances with agreeing labels. We use the generalized cross entropy loss of Zhang & Sabuncu (2018).

(xi,ℓi,ei) L

log Peiφ(reii = 1|xi) + X

j:xi Hj ℓi =ℓj log Pjφ(rji = 0|xi)

j:xi Hj ℓi=ℓj Generalized-XENT(Pjφ(rj|xi), rji = 1) (2)

Note for other instances xi in Rj s cover Hj, value of rji is unknown and latent. The second type of supervision is on the relationship between rji and yi for each xi Hj. A rule Rj imposes a causal constraint that when rji = 1, the label yi has to be ℓj. rji = 1 = yi = ℓj xi Hj (3) We convert this hard constraint into a (log) probability of the constraint being satisfied under the Pθ(y|x) and Pjφ(rj|x) distributions as:

log 1 Pjφ(rj = 1|x)(1 Pθ(ℓj|x)) (4)

Figure 2 shows a surface plot of the above log probability as a function of Pθ(ℓj|x) (shown as axis P(y) in figure) and Pjφ(rj = 1|x) (shown as axis P(r) in figure) for a single rule.

Published as a conference paper at ICLR 2020

Figure 2: Negative implication loss

Observe that likelihood drops sharply as P(rj|x) is close to 1 but P(y = ℓj|x) is close to zero. For all other values of these probabilities the log-likelihood is flat and close to zero. Specifically, when Pjφ predicts low values of rj for a x, the log-likelihood surface is flat, effectively withdrawing the (x, ℓj) supervision from training the classifier Pθ. Thus maximizing this likelihood provides a soft enforcement of the constraint without unwanted biases. We call this the negative implication loss.

We do not need to explicitly model the conflict among rules, that is when an xi is covered by two rules Rj and Rk of differing labels (ℓj = ℓk), then both rji and rki cannot be 1. This is because the constraint among pairs (yi, rji) and (yi, rki) as stated in Equation 3 subsumes this one.

During training we then seek to maximize the log of the above probability along with normal data likelihood terms. Putting the terms in Equations 1, 2 and 4 together our final training objective is:

min θ,φ LL(θ) LL(φ) γ X

j;x Hj U log(1 Pjφ(rj = 1|x)(1 Pθ(ℓj|x))) (5)

We refer to our training loss as a denoised rule-label implication loss or Imply Loss for short. The LL(φ) term seeks to denoise rule coverage which then influence the y distribution via the implication loss. We explored several other methods of enforcing the constraint among y and rj in the training of the Pθ and Pjφ networks. Our method Imply Loss consistently performed the best among several methods we tried including the recent posterior regularization (Ganchev et al., 2010; Hu et al., 2016) method of enforcing soft constraints and co-training (Blum & Mitchell, 1998).

Network Architecture Our network has three modules. (1) A shared embedding layer that provides the feature representation of the input. When labeled data is scarce, this will typically be a pre-trained layer from a related task. The embedding module is task-specific and is described in the experiment section. (2) A classification network that models Pθ(y|x) with parameters θ. The embedding of an input x is passed through multiple non-linear layers with Re LU activation, a last linear layer followed by Softmax to output a distribution over the class labels. (3) A rule network that models Pjφ(rj = 1|x) whose parameters φ are shared across all rules. The input to the network is rule-specific and concatenates the embedding of the input instance x, and a one-hot encoding of the rule id j . The input is passed through multiple non-linear layers with Re LU activation before passing through a Sigmoid activation which outputs the probability Pjφ(rj = 1|x).

Inference During prediction, joint inference over the label y and coverage variables rj provides slight gains over depending solely on Pθ(y|x). For any test example x, consider the set of rules G covering x such that Pjφ(1|x) > 0.5. Probabilities from the label and coverage variables are combined to obtain a score s(y) for each label y as:

s(y|x) = Pθ(y|x) +

Rj G δ(ℓj = y)Pjφ(1|x) + δ(ℓj = y)Pjφ(0|x)

The above can be viewed as a soft voting over the trained classifier Pθ and labels provided by rules with uncertain coverage. Because we also learned to denoise rules along with training the classifier, the labels assigned by the rules have higher precision than original rules.

3 EXPERIMENTS

We compare our training algorithms against simple baselines, existing error-tolerant learning algorithms, and existing constraint-based learning in deep networks.

We evaluate across five datasets spanning three task types: text classification, sequence labeling, and record classification. We augment the datasets with rules, that we obtained manually in three

Published as a conference paper at ICLR 2020

Dataset |L| |U| #Rules %Cover Precision %Conflict Avg |Hj|

#Rules Per Instance

|Valid| |Test|

Question 68 4884 68 95 63.8 22.5 124 1.8 500 500 MIT-R 1842 64888 15 14 80.7 2.5 634 1.1 4091 14256 SMS 69 4502 73 40 97.3 0.6 31 1.3 500 500 You Tube 100 1586 10 87 78.6 30.2 258 1.9 120 250 Census 83 10000 83 100 84.1 27.5 540 4.5 5561 16281 Table 1: Statistics of datasets and their rules. %Cover is fraction of instances in U covered by at least one rule. Precision refers to micro precision of rules. Conflict denotes the fraction of instances covered by conflicting rules among all the covered instances. Avg |Hj| is average cover size of a rule in U. Rules Per Instance is average number of rules covering an instance in U.

cases, from pre-existing public sources in one case, and automatically in another. Table 1 presents statistics summarizing the datasets and rules. A brief description of each appears below.

Question Classification (Li & Roth, 2002): This is a TREC-6 dataset to classify a question to one of six categories: {Abbreviation, Entity, Description, Human, Location, Numeric-value}. The training set has 5452 instances which are split as 68 for L, 500 for validation, and the remaining as U. Each example in L is generalized as a rule represented by a regular expression. E.g. After labeling How do you throw a housewarming party ? as Description we define a rule (how|How|what|What)(does|do|to|can). Description. More rules in Table 4 of supplementary. Although, creating such 68 generalised rules required 90 minutes, the generalizations cover 4637 instances in U, almost two orders of magnitude more instances than in L! On an average each of our rule covered 124 instances (|Hj| column in Table 1). But the precision of labels assigned by rules was only 63.8%. 22.5% of covered instances had an inter-rule conflict, demonstrating noise in the rule labelings. Accuracy is used as the performance metric.

MIT-R1 (Liu et al., 2013): This is a slot-filling task on sentences about restaurant search and the task is to label each token as one of {Location, Hours, Amenity, Price, Cuisine, Dish, Restaurant Name, Rating, Other}. The training data is randomly split into 200 sentences (1842 tokens) as L, 500 sentences (4k tokens) as validation and remaining 6.9k sentences (64.9k tokens) as U. We manually generalize 15 examples in L. E.g. After inspecting the sentence where can i get the highest rated burger within ten miles and labeling highest rated as Rating, we provide the rule: . (highly|high|good|top|highest)(rate|rating|rated). Rating to the matched positions. More examples in Table 7 of supplementary. Although, creating 15 generalizing rules took 45 minutes of annotator effort, the rules covered roughly 9k tokens in U. F1 metric is used for evaluation on the default test set of 14.2k tokens over 1.5k sentences.

SMS Spam Classification (Almeida et al., 2011): This dataset contains 5.5k text messages labeled as spam/not-spam, out of which 500 were held out for validation and 500 for testing. We manually generalized 69 exemplars to rules. Remaining examples go in the U set. The rules here check for presence of keywords or phrases in the SMS .* guaranteed gift .* spam. A rule covers 31 examples on an average and has a precision of 97.3%. However, in this case only 40% of the unlabeled set is covered by a rule. We report F1 here since class is skewed. More examples in Table 5 of supplementary.

Youtube Spam Classification (Alberto et al., 2015): Here the task is to classify comments on You Tube videos as Spam or Not-Spam. We obtain this from Snorkel s Github page2, which provides 10 labeling functions which we use as rules, an unlabeled train set which we use as U, a labeled dev set to guide the creation of their labeling functions which we use as L, and labeled test and validation sets which we use in the same roles. Their labeling functions have a large coverage (258 on average), and a precision of 78.6%.

Census Income (Dua & Graff, 2019): This UCI dataset is extracted from the 1994 U.S. census. It lists a total of 13 features of an individual such as age, education level, marital status, country of

1groups.csail.mit.edu/sls/downloads/restaurant/ 2https://github.com/snorkel-team/snorkel-tutorials/tree/master/spam

Published as a conference paper at ICLR 2020

Methods Datasets Question (Accuracy)

You Tube (Accuracy)

Census (Accuracy) Majority (No parameters trained) 60.9 (0.7) 40.9 (0.1) 82.2 (0.9) 48.4 (1.2) 80.1 (0.1) Only-L 72.9 (0.6) 73.5 (0.3) 90.9 (1.8) 89.0 (1.6) 79.4 (0.5)

L+Umaj - 1.4 (1.5) + 0.0 (0.3) + 0.8 (1.9) + 3.5 (1.2) + 0.9 (0.1) Noise-tolerant (Zhang et al., 2018) - 0.5 (1.1) + 0.0 (0.2) + 1.7 (1.1) + 2.9 (1.2) + 1.0 (0.2) L2R (Ren et al., 2018b) + 0.3 (2.1) - 15.4 (1.0) + 2.5 (0.5) + 2.3 (0.8) + 2.9 (0.3) L+Usnorkel (Ratner et al., 2016) - 0.7 (3.0) + 0.0 (0.2) + 2.7 (0.7) + 3.5 (1.3) + 1.0 (0.4) Snorkel-Noise-Tolerant - 1.4 (1.6) + 0.0 (0.3) + 2.0 (0.7) + 2.7 (1.5) + 0.2 (0.5) Posterior Reg. (Hu et al., 2016) - 0.8 (1.0) - 0.1 (0.4) - 2.9 (1.9) + 1.8 (1.5) - 0.8 (0.5) Imply Loss (Ours) + 11.7 (1.5) + 0.8 (0.3) + 3.2 (1.1) + 4.2 (1.0) + 1.7 (0.2) Table 2: Comparison of Imply Loss (our method) with various methods (described in Section 3.1) on five different datasets. The numbers reported for all methods after the double-line are gains over the baseline (Only L) that does not use rules at all. Higher is better. NOTE: Numbers in brackets represent standard deviation of the original accuracy and not of gains.

origin etc. The primary task on it is binary classification - whether a person earns more than $50K or not. The train data consists of 32563 records. We choose 83 random data points as L, 10k points as U and 5561 points as validation data. For this case we created the rules synthetically as follows: We hold out disjoint 16k random points from the training dataset as a proxy for human knowledge and extract a PART decision list (Frank & Witten, 1998) from it as our set of rules. We retain only those rules which fire on L.

Network Architecture Since our labeled data is small we depend on pre-trained resources. As the embedding layer we use a pretrained ELMO (Peters et al., 2018) network where 1024 dimensional contextual token embeddings serve as representations of tokens in the MIT-R sentences, and their average serve as representation for sentences in Question and SMS dataset. Parameters of the embedding network are held fixed during training. For sentences in the You Tube dataset, we use Snorkel s2 architecture of a simple bag-of-words feature representation marking the frequent unigrams and bi-grams present in a sentence using a few-hot vector. For the Census dataset categorical features are represented as one hot vectors, while real valued features are simply normalized. For MIT-R, Question and SMS both classification and rule-weight network contain two 512 dimensional hidden layers with Re LU activation. For Census, both the networks contain two 256 dimensional hidden layers with Re LU activation. For You Tube, the classifier network is a simple logistic regression like in Snorkel s code. The rule network has one 32-dimensional hidden layer with Re LU activation.

Each reported number is obtained by averaging over ten random initializations. Whenever a method involved hyper-parameters to weigh the relative contribution of various terms in the objective, we used a validation dataset to tune the value of the hyper-parameter. Hyperparameters used are provided in Section C of supplementary.

3.1 COMPARISON WITH DIFFERENT METHODS In Table 2 we compare our method with the following alternatives on each of the five datasets:

Majority: that predicts via majority vote among the rules that cover an instance. This baseline indicates the stand-alone quality of rules, no network is learned here. Ties are broken arbitrarily for class-balanced datasets or by using a default class. Table 2, shows that the accuracy of majority is quite poor indicating either poor precision or poor coverage of the rule sets.3.

Only-L: Here we train the classifier Pθ(y|x) only on the labeled data L using the standard crossentropy loss (Equation 1). Rule generalisations are not utilized at all in this case. We observe in Table 2 that even with the really small labeled set we used for each dataset, the accuracy of a classifier learned with clean labeled data is much higher than noisy majority labels of rules. We consider this method as our baseline and report the gains on remaining methods.

3Only for the Census dataset the relative accuracy is high because the rules were obtained synthetically through a rule-learning algorithm on a very large labeled dataset to serve as a proxy for a human s generalization.

Published as a conference paper at ICLR 2020

L+Umaj: Next we train the classifier on L along with Umaj obtained by labeling instances in U with the majority label among the rules applicable to the instance. Loss corresponding to the examples labeled by rules is weighted as follows:

(xj,ℓj) L log Pθ(ℓj|xj) + γ X

(xj,yj) Umaj log Pθ(yj|xj) (7)

The row corresponding to L+Umaj in Table 2 provides the gains of this method over Only-L. We observe gains with the noisily labeled U in three out of the five cases.

Noise-tolerant: Since labels in Umaj are noisy, we next use Zhang & Sabuncu (2018) s noise tolerant generalized cross entropy loss on them with regular cross-entropy loss on the clean L as follows:

(xj,ℓj) L log Pθ(ℓj|xj) + γ X

(xj,yj) Umaj

(1 Pθ(yj|x))q

Parameter q [0, 1] controls the noise tolerance which we tune as a hyper-parameter. We observe that in three cases minimizing the above objective improves beyond L+Umaj validating that noisetolerant loss functions can be useful for learning from noisy labels on Umaj.

Learning to Reweight (L2R) (Ren et al., 2018b): is a recent method for training with a mix of clean and noisy labeled data. They train the classifier by meta-learning to re-weight the loss on the noisily labelled instances (Umaj) with the help of the clean examples (L). This method provides significant accuracy gains over Only-L in three out the five datasets. However, it fails in the multiclass classification task of slot-filling which has a very high class imbalance and rules of smaller coverage.

All the above methods employ no extra parameters to denoise or weight individual rules. We next compare with a number of methods that do.

L+Usnorkel: This method replaces Majority-based consensus with Snorkel s generative model (Ratner et al., 2016) that assigns weights to rules and labels examples in U. Thereafter we use the same approach as in L+Umaj with just Snorkel s soft-labels instead of Majority on U. We also compare with using noise-tolerant loss on U labeled by Snorkel (Eqn:8) which we call Snorkel Noise-Tolerant. Like previous methods, both of these methods provide improvements over Only-L on three of the five datasets where the rules are less noisy. L+Usnorkel performs slightly better than Noise-Tolerant on Umaj.

We next compare with a method that simultaneously learns two sets of networks Pθ and Pjφ like ours but with different loss function and training schedule.

Posterior Regularization (PR): This method proposed in Hu et al. (2016) also treats rules as softconstraints and has been used for training neural networks for structured outputs. They use Ganchev et al. (2010) s posterior regularization framework to train the two networks in a teacher-student setup. We adapt the same framework and get a procedure as follows: The student proposes a distribution over y and rjs using current Pθ and Pjφ, the teacher uses the constraint in Eq 3 to revise the distributions so as to minimize the probability of violations, the student updates parameters θ and φ to minimize KL distance with the revised distribution. The detailed formulation appear in the Section A of supplementary. We find that this method is no better than Only-L in most of the cases and worse than the noise-tolerant method that does not train extra φ parameters.

Imply Loss(Ours): Overall our approach of training with denoised rule-label implication loss provides much better accuracy than all the above eight methods and we get consistent gains over Only-L on all datasets. On the Question dataset we get 11.7 points gain over Only-L whereas the best gain by existing method was 0.3. A useful property of our method compared to the PR method above is that the training process is simple and fits into the batch stochastic gradient training template. In contrast, PR requires special alternating computations. We next perform a number of diagnostics experiments to explain the reasons for the superior performance of our method.

Published as a conference paper at ICLR 2020

Question MIT-R You Tube SMS Census

Old precision Denoised Precision Percent Suppressed

Figure 3: Rule-specific denoising by our method.

Diagnostics: Effectiveness of learning true coverage via Pjφ An important part of our method is the rule-specific denoising learned via the Pjφ network. In the chart alongside we plot the original precision of rules on the test data, and the precision after suppressing those rule labelings where Pjφ(rj|x) predicts 0 instead of 1. Observe now that the precision is more than 91% on all datasets. For the Question dataset, the precision jumped from 64% to 98%. The percentage of labelings suppressed (shown by the dashed line) is higher on datasets with noisier rules (e.g. compare Question and SMS). This shows that Pjφ is able to denoise rules by capturing the distribution of the latent true coverage variables with the limited LL(φ) loss and indirectly via the implication loss.

Rule Precision

56 66 71 75 83

Figure 4: Effect of rule precision

Effect of rule precision Rules in the Census dataset are of higher quality in terms of precision as well as coverage. Superior performance of the L2R method on this dataset motivated us to inspect how well our method performs on the same dataset in the absence of high precision rules. We created four new versions of the rule sets by successively removing high precision rules from the original rule set. We observe that our method performs better than L2R when rules have low precision. Because Imply Loss denoises rules, it is better able to handle low-precision rules.

Role of Exemplars in Rules We next evaluate the importance of the exemplar-rule pairs in learning the Pjφ and Pθ networks. The exemplars of a rule give an interesting new form of supervision about an instance where a labeling rule must fire. To evaluate the importance of this supervision, we exclude the rj = 1 likelihood on rule-exemplar pairs from LL(φ), that is, the first term in Equation 2 is dropped. In the table below we see that performance of Imply Loss usually drops when the exemplar-rule supervision is removed. Interestingly, even after this drop, the performance of Imply Loss surpasses most of the methods in Table 2 indicating that even without exemplar-rule pairs our training objective is effective in learning from rules and labeled instances.

Question MIT-R SMS Census rj = 1 for rule-exemplar pairs 84.5 (1.5) 73.7 (0.3) 93.2 (1.0) 81.0 (0.2) No rj = 1 for rule-exemplar pairs 83.8 (0.7) 73.5 (0.5) 93.5 (1.2) 80.8 (0.3) Table 3: Effect of removing rule-exemplar supervision from LL(φ)

68 200 400 600 800

Ours Posterior Reg. L+Usnorkel Only-L

Figure 5: Effect of increasing labeled data

Effect of increasing labeled data L We increase L while keeping the number of rules fixed on the Question dataset. In the attached plot we see the accuracy of our method (Imply Loss) against Only-L, L+Usnorkel and Posterior Reg. We observe the expected trend that the gap between the method narrows as labeled data increases.

4 RELATED WORK Learning from noisily labeled data has been extensively studied in settings like crowdsourcing. One category of these algorithms upper-bound the loss function to make it robust to noise. These include methods like MAE (Ghosh

Published as a conference paper at ICLR 2020

et al., 2017), Generalized Cross Entropy (CE)(Zhang & Sabuncu, 2018), and Ramp loss (Collobert et al., 2006). Most of these assume that noise is independent of the input given the true label. In our model noise is systematic and instance-dependent.

A second category assume that a small clean dataset is available along with noisily labeled data. This is also true in our case, and we compared with a state of the art method in that category Ren et al. (2018b) that chooses a descent direction that aligns with a clean validation set using metalearning. Others in this category include: Shen & Sanghavi (2019) s method of iteratively selecting examples with smallest loss, and Veit et al. (2017) s method of learning a separate network to transform noisy labels to cleaned ones which are used to impose a cross-entropy loss on Pθ(y|x). In contrast, we perform rule-specific cleaning via latent coverage variables and a flexible implication loss which withdraws y supervision when Pjφ(rji|x) assumes low values. Another way of relating clean and noisy labels is via an instance-independent confusion matrix learned jointly with the classifier (Khetan et al., 2018; Goldberger & Ben-Reuven, 2016; Han et al., 2018b;a). These works assume that the confusion matrix is instance independent, which does not hold for our case. Tanaka et al. (2018) uses confidence from the classifier to eliminate noise but they need to ensure that the network does not memorize noise. Our learning setup also has the advantage of extracting confidence from a different network. There is growing interest in integrating logical rules with labeled examples for training networks, specifically for structured outputs (Manhaeve et al., 2018; Xu et al., 2018; Fischer et al., 2019; Sun et al., 2018; Ren et al., 2018a). Xu et al. (2018); Fischer et al. (2019) convert rules on output nodes of network, to (almost differentiable) loss functions during training. The primary difference of these methods from ours is that they assume that rules are correct whereas we assume them to be noisy. Accordingly, we simultaneously correct the rules and use them to improve the classifier, whereas they use the rules as-is to train the network outputs.

A well-known framework for working with soft rules is posterior regularization (Ganchev et al., 2010) which is used in Hu et al. (2016) to train deep structured output networks while harnessing logic rules. Ratner et al. (2016) works only with noisy rules treating them as black-box labeling functions and assigns a linear weight to each rule based on an agreement objective. Our learning model is more powerful that attempts to learn a non-linear network to restrict rule boundaries rather than just weight their outputs. We presented a comparison with both these approaches in the experimental section, and showed superior performance.

To the best of our knowledge, our proposed paradigm of coupled rule-exemplar supervision is novel, and our proposed training algorithm is able to harness them in ways not possible by existing frameworks for learning from rules or noisy supervision.

5 CONCLUSION We proposed a new rule-exemplar model for collecting human supervision to combine the scalability of top-level rules with the quality of instance-level labels. We show that such supervision is natural since humans typically inspect examples to code rules. Furthermore, such coupled examples provide supervision on correct firing of rules which help to denoise rules. We propose to train the classifier while jointly denoising rules via latent coverage variables imposing a soft-implication constraint on the true label. Empirically on five datasets we show that our training algorithm that performs rule-specific denoising is better than generic noise-tolerant learning. In future we plan to deploy this framework on other applications where human supervision is a scarce resource.

Reproducibility Code and Data for the experiments available at https://github.com/awasthiabhijeet/Learning-From-Rules

Acknowledgements We thank the anonymous reviewers for their constructive feedback on this work. This research was partly sponsored by a Google India AI/ML Research Award and partly by the IBM AI Horizon Networks - IIT Bombay initiative. Abhijeet is supported by Google Ph D Fellowship in Machine Learning.

Published as a conference paper at ICLR 2020

T ulio C Alberto, Johannes V Lochter, and Tiago A Almeida. Tubespam: Comment spam filtering on youtube. In 2015 IEEE 14th International Conference on Machine Learning and Applications (ICMLA), pp. 138 143. IEEE, 2015.

Tiago A Almeida, Jos e Mar ıa G Hidalgo, and Akebo Yamakami. Contributions to the study of sms spam filtering: new collection and results. In Proceedings of the 11th ACM symposium on Document engineering, pp. 259 262. ACM, 2011.

Douglas E. Appelt, Jerry R. Hobbs, John Bear, David J. Israel, and Mabry Tyson. Fastus: A finitestate processor for information extraction from real-world text. In IJCAI, pp. 1172 1178, 1993.

Stephen H. Bach, Daniel Rodriguez, Yintao Liu, Chong Luo, Haidong Shao, Cassandra Xia, Souvik Sen, Alexander Ratner, Braden Hancock, Houman Alborzi, Rahul Kuchhal, Christopher R e, and Rob Malkin. Snorkel drybell: A case study in deploying weak supervision at industrial scale. In Proceedings of the 2019 International Conference on Management of Data, SIGMOD Conference 2019, Amsterdam, The Netherlands, June 30 - July 5, 2019., pp. 362 375, 2019.

Avrim Blum and Tom Mitchell. Combining labeled and unlabeled data with co-training. In COLT, 1998.

R. Collobert, F. Sinz, J. Weston, and L. Bottou. Trading convexity for scalability. In ICML 2006, 2006.

Hamish Cunningham. Gate: A framework and graphical development environment for robust nlp tools and applications. In Proc. 40th Annual Meeting of the Association for Computational Linguistics (ACL 2002), pp. 168 175, 2002.

Arthur P Dempster, Nan M Laird, and Donald B Rubin. Maximum likelihood from incomplete data via the em algorithm. Journal of the Royal Statistical Society: Series B (Methodological), 39(1): 1 22, 1977.

Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. ar Xiv preprint ar Xiv:1810.04805, 2018.

Dheeru Dua and Casey Graff. UCI machine learning repository, 2019. URL http://archive.ics.uci. edu/ml.

Marc Fischer, Mislav Balunovic, Dana Drachsler-Cohen, Timon Gehr, Ce Zhang, and Martin Vechev. DL2: Training and querying neural networks with logic. In Proceedings of the 36th International Conference on Machine Learning, pp. 1931 1941, 2019.

Eibe Frank and Ian H. Witten. Generating accurate rule sets without global optimization. In J. Shavlik (ed.), Fifteenth International Conference on Machine Learning, pp. 144 151. Morgan Kaufmann, 1998.

Kuzman Ganchev, Jennifer Gillenwater, Ben Taskar, et al. Posterior regularization for structured latent variable models. Journal of Machine Learning Research, 11(Jul):2001 2049, 2010.

Aritra Ghosh, Himanshu Kumar, and PS Sastry. Robust loss functions under label noise for deep neural networks. In Thirty-First AAAI Conference on Artificial Intelligence, 2017.

Garrett B. Goh, Charles Siegel, Abhinav Vishnu, and Nathan Hodas. Using rule-based labels for weak supervised learning: A chemnet for transferable chemical property prediction. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD 18, 2018.

Jacob Goldberger and Ehud Ben-Reuven. Training deep neural-networks using a noise adaptation layer. 2016.

Bo Han, Jiangchao Yao, Gang Niu, Mingyuan Zhou, Ivor Tsang, Ya Zhang, and Masashi Sugiyama. Masking: A new perspective of noisy supervision. In Advances in Neural Information Processing Systems, pp. 5841 5851, 2018a.

Published as a conference paper at ICLR 2020

Bo Han, Quanming Yao, Xingrui Yu, Gang Niu, Miao Xu, Weihua Hu, Ivor Tsang, and Masashi Sugiyama. Co-teaching: Robust training of deep neural networks with extremely noisy labels. In Advances in Neural Information Processing Systems 31, pp. 8536 8546. 2018b.

Zhiting Hu, Xuezhe Ma, Zhengzhong Liu, Eduard Hovy, and Eric Xing. Harnessing deep neural networks with logic rules. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), August 2016.

Dongyeop Kang, Tushar Khot, Ashish Sabharwal, and Eduard Hovy. Adventure: Adversarial training for textual entailment with knowledge-guided examples. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Association for Computational Linguistics, 2018.

Ashish Khetan, Zachary C. Lipton, and Anima Anandkumar. Learning from noisy singly-labeled data. In International Conference on Learning Representations, 2018. URL https://openreview. net/forum?id=H1s UHgb0Z.

Xin Li and Dan Roth. Learning question classifiers. In Proceedings of the 19th international conference on Computational linguistics-Volume 1, pp. 1 7. Association for Computational Linguistics, 2002.

Jingjing Liu, Panupong Pasupat, Yining Wang, Scott Cyphers, and Jim Glass. Query understanding enhanced by hierarchical parsing structures. In 2013 IEEE Workshop on Automatic Speech Recognition and Understanding, pp. 72 77. IEEE, 2013.

Robin Manhaeve, Sebastijan Dumancic, Angelika Kimmig, Thomas Demeester, and Luc De Raedt. Deepproblog: Neural probabilistic logic programming. In Advances in Neural Information Processing Systems 31, pp. 3749 3759. 2018.

Arghya Pal and Vineeth N. Balasubramanian. Adversarial data programming: Using gans to relax the bottleneck of curated labeled data. In 2018 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2018, Salt Lake City, UT, USA, June 18-22, 2018, pp. 1556 1565, 2018.

Matthew E Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kenton Lee, and Luke Zettlemoyer. Deep contextualized word representations. ar Xiv preprint ar Xiv:1802.05365, 2018.

Alexander J Ratner, Christopher M De Sa, Sen Wu, Daniel Selsam, and Christopher R e. Data programming: Creating large training sets, quickly. In Advances in Neural Information Processing Systems 29. 2016.

Hongyu Ren, Russell Stewart, Jiaming Song, Volodymyr Kuleshov, and Stefano Ermon. Learning with weak supervision from physics and data-driven constraints. AI Magazine, 39(1):27 38, 2018a.

Mengye Ren, Wenyuan Zeng, Bin Yang, and Raquel Urtasun. Learning to reweight examples for robust deep learning. ar Xiv preprint ar Xiv:1803.09050, 2018b.

Yanyao Shen and Sujay Sanghavi. Learning with bad training data via iterative trimmed loss minimization. In Proceedings of the 36th International Conference on Machine Learning, pp. 5739 5748, 2019.

Chen Sun, Abhinav Shrivastava, Saurabh Singh, and Abhinav Gupta. Revisiting unreasonable effectiveness of data in deep learning era. In IEEE International Conference on Computer Vision, ICCV 2017, Venice, Italy, October 22-29, 2017, pp. 843 852, 2017.

Haitian Sun, William W Cohen, and Lidong Bing. Semi-supervised learning with declaratively specified entropy constraints. In Advances in Neural Information Processing Systems 31, pp. 4425 4435. 2018.

Daiki Tanaka, Daiki Ikami, Toshihiko Yamasaki, and Kiyoharu Aizawa. Joint optimization framework for learning with noisy labels. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 5552 5560, 2018.

Published as a conference paper at ICLR 2020

Michael Henry Tessler and Noah D. Goodman. The language of generalization. Psychological Review, 126(3):395 436, 2019.

Andreas Veit, Neil Alldrin, Gal Chechik, Ivan Krasin, Abhinav Gupta, and Serge Belongie. Learning from noisy large-scale datasets with minimal supervision. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 839 847, 2017.

Jingyi Xu, Zilu Zhang, Tal Friedman, Yitao Liang, and Guy Van den Broeck. A semantic loss function for deep learning with symbolic knowledge. In Proceedings of the 35th International Conference on Machine Learning, pp. 5502 5511, 2018.

Zhilu Zhang and Mert Sabuncu. Generalized cross entropy loss for training deep neural networks with noisy labels. In Advances in Neural Information Processing Systems 31. 2018.

Published as a conference paper at ICLR 2020

Supplementary Material: Learning from Rules Generalizing Labeled Exemplars

A POSTERIOR REGULARIZATION METHOD

We model a joint distribution Q(y, r1, . . . , rn|x) to capture the interaction among the label random variable y and coverage random variables r1, . . . , rn of any instance x. We use r to compactly represent r1, . . . , rn. Strictly speaking, when a rule Rj does not cover x, the rj is not a random variable and its value is pinned to 0 but we use this fixed-tuple notation for clarity. The random variables rj and y impose a constraint on the joint distribution Q: for a x Hj when rj = 1, the label y cannot be anything other than ℓj. rj = 1 = y = ℓj x Hj (9) We can convert this into a soft constraint on the marginals of the distribution Q by stating the probability of P

y =ℓj Q(y, rj = 1|x) should be small.

y =ℓj Q(y, rj = 1|x) (10)

The singleton marginals of Q along the y and rj variables are tied to the Pθ and Pjφ(rj|x) we seek to learn. A network with parameters θ models the classifier Pθ(y|x), and a separate network with φ variables (shared across all rules) learns the Pjφ(rj|x) distribution. The marginals of joint Q should match these trained marginals and we use a KL term for that:

KL(Q(y|x); Pθ(y|x)) + X

j:x Hj KL(Q(rj|x); Pjφ(rj|x)) (11)

We call the combined KL term succinctly as KL(Q, Pθ) + KL(Q, Pφ).

Further the Pθ and Pjφ distributions should maximize the log-likelihood on their respective labeled data as provided in Equation 1 and Equation 2 respectively.

Putting all the above objectives together with hyper-parameters α > 0, λ > 0 we get our final objective as:

min Q,θ,φ α(LL(θ) + LL(φ)) + KL(Q, Pθ) + KL(Q, Pφ) + λ X

y =ℓj Q(y, rj = 1|x) (12)

We show in Section A.1 that this gives rise to the solution for Q in terms of Pθ, Pjφ and alternately for Pθ, Pjφ in terms of Q as follows.

Q(y, r|x) Pθ(y|x) Y

j:x Hj Pjφ(rj|x)e λδ(y =ℓj rj=1) (13)

where δ(y = ℓj rj = 1) is an indicator function that is 1 when the constraint inside holds, else it is 0. Computing marginals of the above using straight-forward message passing techniques we get:

Q(y|x) Pθ(y|x) Y

j:x Hj (Pjφ(1|x)e λδ(y =ℓj) + Pjφ(0|x)) (14)

Q(rk = 1|x) Pkφ(1|x) X

y e λδ(y =ℓk)Pθ(y|x) Y

j =k,x Hj (Pjφ(1|x)e λδ(y =ℓj) + Pjφ(0|x))

(15) Thereafter, we solve for θ and φ in terms of a given Q as

min θ,φ LL(θ) LL(φ) γ X

y Y Q(y|xi) log Pθ(y|xi) + X

rj {0,1} Q(rj|xi) log Pjφ(rj|xi)

Here, γ = 1 α. This gives rise to an alternating optimization algorithm as in the posterior regularization framework of Ganchev et al. (2010). We initialize θ and φ randomly. Then in a loop, we perform the following two steps alternatively much like the EM algorithm (Dempster et al., 1977).

Published as a conference paper at ICLR 2020

Q Computation step: Here we compute marginals Q(y|x) and Q(rj|x) from current Pθ and Pjφ using Equations 14 and 15 respectively for each x in a batch. This computation is straight-forward and does not require any neural optimization. We can interpret the Q(y|x) as a small correction of the Pθ(y|x) so as to align better with the constraints imposed by the rules in Equation 3. Likewise Q(rj|x) is an improvement of current Pjφs in the constraint preserving direction. For example, the expected rj values might be reduced for an instance if its probability of y being ℓj is small.

Parameter update step: We next reoptimize the θ and φ parameters to match the corrected Q distribution as shown in Equation 16. This is solved using standard stochastic gradient techniques. The Q terms can just be viewed as weights at this stage which multiply the loss or label likelihood. A pseudocode of our overall training algorithm is described in Algorithm 1.

Algorithm 1 Our Joint Training Algorithm using Posterior Regularization

Input: L, U

Initialize parameters θ, φ randomly for a random training batch from U L do

Obtain Pθ(y|x) from the classification network. Obtain Pjφ(rj|x)j [n] from the rule-weight network. Calculate Q(y|x) using Eqn 14 and Q(rj|x)j [n] using Eqn 15. Update θ and φ by taking a step in the direction to minimize the loss in Eqn 16. end for Output: θ , φ

A.1 PROOF: ALTERNATING SOLUTION FOR OPTIMIZATION OBJECTIVE IN EQN 12

Treat each Q(y, r) as an optimization variable with the constraint that P y,r Q(y, r) = 1. We express this constraint with a Langrangian multiplier η in the objective. Also, define a distribution

Pθ,φ(y, r|x) = Pθ(y|x) Y

j:x Hj Pjφ(rj|x)

It is easy to verify that the KL terms in our objective 12 can be collapsed as KL(Q; Pθ,φ). The rewritten objective (call it F(Q, θ, φ) ) is now:

α(LL(θ) + LL(φ)) + X

x KL(Q(y, r|x), Pθ,φ(y, r|x))

y =ℓj Q(y, rj = 1|x) + η(1 X

y,r Q(v, r)) (17)

Next we solve for F Q(y,r) = 0 after expressing the marginals in their expanded forms: e.g. Q(y, rj|x) = P

r1,...,rj 1,rj+1,...,rn Q(y, r1, . . . , rn|x). This gives us

F Q(y, r) = log Q(y, r) log Pθ,φ(y, r|x)

j:x Hj λδ(y = ℓj, rj = 1) + η + 1

Equating it to zero and substituting for Pθ,φ we get the solution for Q(y, r) in Equation 13.

The proof for the optimal Pθ and Pjφ while keeping Q fixed in Equation 17 is easy and we skip here.

Published as a conference paper at ICLR 2020

B LIST OF RULES

We provide a list of rules for each task type.

Rule Example Class ( |ˆ)(where)[ˆ\w]* (\w+ ){0,1} (was|is)[ˆ\w]*( |\$)

Where is Trinidad ? Location

( |ˆ)(which|what)[ˆ\w]* (\w+ ){0,1} (play|game|movie|book)[ˆ\w]*( |$)

What book is the follow-up to Future Shock ?

( |ˆ)(what)[ˆ\w]* (\w+ ){0,1} (part|division|ratio|percentage) [ˆ\w]*( |$)

Of children between the ages of two and eleven , what percentage watch The Simpsons ?

( |ˆ)(who|who)[ˆ\w]* (\w+ ){0,1} (found|discovered|made|built |build|invented)[ˆ\w]*( |$)

Who invented volleyball ? Human

Table 4: Sample rules for TREC Question Classification. Rule fires if the regex matches

Rule Example Class ( |ˆ)(free)[ˆ\w]* ([ˆ\s]+ )*(price)[ˆ\w]* ([ˆ\s]+ )*(call)[ˆ\w]*( |$)

Free video camera phones with Half Price line rental for 12 mths and 500 cross ntwk mins 100 txts. Call Mobile Upd8 08001950382 or Call2Opt Out/674

( |ˆ)(guranteed)[ˆ\w]* ([ˆ\s]+ )* (gift\.|gift)[ˆ\w]*( |$)

Great News! Call FREEFONE 08006344447 to claim your guaranteed a 1000 CASH or a 2000 gift.

( |ˆ)(can t)[ˆ\w]* (\w+ ){0,1}(talk)[ˆ\w]*( |$)

sry can t talk on phone, with parents Not Spam

( |ˆ)(that s)[ˆ\w]* (\w+ ){0,1}(fine!|fine)[ˆ\w]*( |$)

Yeah, that s fine! It s a 6 to get in, is that ok?

Table 5: Sample rules for Spam Classification. Rule fires if the regex matches

Rules Class capital-gain > 6849 > 50K education-num > 12 AND marital-status = Never-married AND native-country = United-States AND occupation = Exec-managerial

marital-status = Separated AND hours-per-week 41

education-num 12 AND native-country = United-States AND age 30

Table 6: Sample rules for census dataset. Rule fires if all clauses are True

Published as a conference paper at ICLR 2020

Rule Example Class ( |ˆ)[ˆ\w]* (within|near|next|close|nearby| around|around)[ˆ\w]*([ˆ\s]+ ){0,2} (here|city|miles|mile) *[ˆ\w]*( |$)

any kid friendly restaurants around here

Word Lists:

cuisine1a=[ italian , american , japanese , spanish , mexican , chinese , vietnamese , vegan ]

cuisine1b=[ bistro , delis ]

cuisine2=[ barbecue , halal , vegetarian , bakery ]

can you find me some chinese food

([0-9]+|few|under [0-9]+) dollar i need a family restaurant with meals under 10 dollars and kids eat

((high|highly|good|best|top| well|highest|zagat) (rate|rating|rated))| ((rated|rate|rating) [0-9]* star)|([0-9]+ star)

where can i get the highest rated burger within ten miles

((open|opened) (now|late))| (still (open|opened|closed|close)) |(((open|close|opened|closed) \w+([\s]| \w* | \w* \w* ))*[0-9]+ (am|pm|((a|p) m)|hours|hour))

where is the nearest italian restaurant that is still open

(outdoor|indoor|group|romantic| family|outside|inside|fine| waterfront|outside|private| business|formal|casual|rooftop| (special occasion)) ([\s]| \w+ | \w+ \w+ )dining

i want to go to a restaurant within 20 miles that got a high rating and is considered fine dining

[\w+ ]{0,2}(palace|cafe|bar| kitchen|outback|dominoes)

is passims kitchen open at 2 am

Restaurant Name

wine|sandwich|pasta|burger| peroggis|burrito| (chicken tikka masala)| appetizer|pizza|wine| cupcake|(onion ring)|tapas

please find me a pub that serves burgers

Table 7: Sample rules for MIT-R dataset. Rule fires if the regex matches or sentence contains a word found in the provided word lists.

Published as a conference paper at ICLR 2020

C HYPERPARAMETERS

Across all experiments we use Adam optimizer with default values of β1, β2, and ϵ. Dropout of 0.8 (keep probability) was used in the feed forward layers. All the models were trained for a maximum of 100 epochs and early stopping was used based on a validation set. Best model on the validation set was evaluated on the test set. Each experiment was run with 10 random initializations. A list of hyperparameters used in our experiments is provided below.

Noisetolerant

Snorkel Noise Tolerant

Post. Reg. implication L+Usnorkel L+Umaj

Question Classification γ 0.001 0.1 0.001 0.1 0.01 0.001 q 0.9 0.6 - - - - lr 0.0003 bs 32 (16 for Only-L)

MIT-R γ 0.01 0.001 0.01 0.1 0.05 0.01 q 0.6 0.6 - - - - lr 0.0003 bs 64 (32 for Only-L)

You Tube γ 0.003 0.5 0.1 0.2 0.5 0.003 q 0.6 0.6 - - - - lr 0.0003 bs 32 (16 for Only-L)

SMS γ 0.1 0.1 0.001 0.3 0.5 0.1 q 0.6 0.6 - - - 0.1 lr 0.0001 bs 32 (16 for Only-L)

Census γ 0.5 0.1 0.001 0.1 0.01 0.5 q 0.1 0.6 - - - 0.5 lr 0.0001 0.0003 bs 64 (16 for Only-L) Table 8: Hyperparameters for various methods and datasets. bs refers to the batch size and lr refers to the learning rate. For Only-L baseline smaller batch size was used considering the smaller size of L set.

Question MIT-R You Tube SMS Census meta lr 0.01 0.0001 0.001 0.0001 0.0001 lr 0.0003 0.0001 0.0003 bs 32 64 32 32 64 Table 9: Meta-learning rate, learning rate and batch size used for L2R (Ren et al., 2018b) for various datasets