United States

Most existing distillation methods ignore the flexible role of the temperature in the loss function and fix it as a hyper-parameter that can be decided by an inefficient grid search. In general, the temperature controls the discrepancy between two distributions and can faithfully determine the difficulty level of the distillation task. Keeping a constant temperature, i.e., a fixed level of task difficulty, is usually sub-optimal for a growing student during its progressive learning stages. In this paper, we propose a simple curriculum-based technique, termed Curriculum Temperature for Knowledge Distillation (CTKD), which controls the task difficulty level during the student&#39;s learning career through a dynamic and learnable temperature. Specifically, following an easy-to-hard curriculum, we gradually increase the distillation loss w.r.t. the temperature, leading to increased distillation difficulty in an adversarial manner. As an easy-to-use plug-in technique, CTKD can be seamlessly integrated into existing knowledge distillation frameworks and brings general improvements at a negligible additional computation cost. Extensive experiments on CIFAR-100, ImageNet-2012, and MS-COCO demonstrate the effectiveness of our method.

AAAI 2023

Curriculum Temperature for Knowledge Distillation

dynamic temperature

computer vision

knowledge distillation

EMNLP 2021 is planned to be a hybrid event in Punta Cana, Dominican Republic, with both on-site and fully virtual participation possible. The experience for on-site participants would closely approximate a normal pre-COVID *ACL conference, with 5-6 thematically organized parallel sessions and live Q/A and interactive discussion immediately after the talks. Presentations by virtual participants will be equitably interleaved with those of on-site participants, projected on the auditorium screens as if on-site, and also followed immediately by live Q/A and interactive discussion at a time during reasonable waking hours for the virtual presenter. For all participants, on-site and virtual, who are unable to attend a session due to either time-zone issues or because they are participating in another session live, talk recordings and slides will be available online at a minimum after the live presentation (and in many cases before as well), and questions may be submitted in advance on session-specific discussion boards and answered live in session with the usual visual aids if desired.

<iframe style="width:700px;height:400px" src="https://online.fliphtml5.com/ebtyf/ceby/"  seamless="seamless" scrolling="no" frameborder="0" allowtransparency="true" allowfullscreen="true" ></iframe>

Please Note: The EMNLP registration system is not currently connected to the underline site as we are still in the process of building out EMNLP 2021. You will receive access instructions from underline the week of November 1st. 

Access is given only to EMNLP upon registration, if you have not registered please do so [here](https://2021.emnlp.org/registration).

Registered attendees will receive access the week of November 1st.

EMNLP 2021

EMNLP 2021 is planned to be a hybrid event in Punta Cana, Dominican Republic, with both on-site and fully virtual participation possible.

Most existing distillation methods ignore the flexible role of the temperature in the loss function and fix it as a hyper-parameter that can be decided by an inefficient grid search. In general, the temperature controls the discrepancy between two distributions and can faithfully determine the difficulty level of the distillation task. Keeping a constant temperature, i.e., a fixed level of task difficulty, is usually sub-optimal for a growing student during its progressive learning stages. In this paper, we propose a simple curriculum-based technique, termed Curriculum Temperature for Knowledge Distillation (CTKD), which controls the task difficulty level during the student's learning career through a dynamic and learnable temperature. Specifically, following an easy-to-hard curriculum, we gradually increase the distillation loss w.r.t. the temperature, leading to increased distillation difficulty in an adversarial manner. As an easy-to-use plug-in technique, CTKD can be seamlessly integrated into existing knowledge distillation frameworks and brings general improvements at a negligible additional computation cost. Extensive experiments on CIFAR-100, ImageNet-2012, and MS-COCO demonstrate the effectiveness of our method.

poster

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-23 is the Thirty-Seventh AAAI Conference on Artificial Intelligence. The theme of this conference is to create collaborative bridges within and beyond AI. Like previous AAAI conferences, AAAI-23 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and two new activities: a Bridge Program and a Lab Program. Many of these activities are tailored to the theme of bridges and all are selected according to the highest standards, with additional programs for students and young researchers.<br><br><br>
AAAI is providing you with a conference planner, which you can use to help organize your itinerary of activities.  This includes talks to attend in person, talks to attend remotely, breaks with colleagues  and your site seeing activities. To access this conference planner, please go to [https://aaai-2023.takemobi.io](https://aaai-2023.takemobi.io).<br><br><br><br>

In order to access this site, you need to register. If you haven't already, please register [here](https://aaai.org/Conferences/AAAI-23/registration/).


The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines.

Only a subset of infections are actually observed in an outbreak, due to multiple reasons such as asymptomatic cases and under-reporting. Therefore, reconstructing an epidemic cascade given some observed cases is an important step in responding to such an outbreak. A maximum likelihood solution to this problem ( referred to as CascadeMLE ) can be shown to be a variation of the classical Steiner subgraph problem, which connects a subset of observed infections. In contrast to prior works on epidemic reconstruction, which consider the standard Steiner tree objective, we show that a solution to CascadeMLE, based on the actual MLE objective, has a very different structure. We design a logarithmic approximation algorithm for CascadeMLE, and evaluate it on multiple synthetic and social contact networks, including a contact network constructed for a hospital. Our algorithm has significantly better performance compared to a prior baseline.

Reconstructing an Epidemic Outbreak Using Steiner Connectivity

Active learning is a critical technique for reducing labelling load by selecting the most informative data. Most previous works applied active learning on Named Entity Recognition (token-level task) similar to the text classification (sentence-level task). They failed to consider the heterogeneity of uncertainty within each sentence and required access to the entire sentence for the annotator when labelling. To overcome the mentioned limitations, in this paper, we allow the active learning algorithm to query subsequences within sentences and propose an Entity-Aware Subsequences-based Active Learning (EASAL) that utilizes an effective Head-Tail pointer to query one entity-aware subsequence for each sentence based on BERT. For other tokens outside this subsequence, we randomly select 30% of these tokens to be pseudo-labelled for training together where the model directly predicts their pseudo-labels. Experimental results on both news and biomedical datasets demonstrate the effectiveness of our proposed method.

EASAL: Entity-Aware Subsequence-based Active Learning for Named Entity Recognition

One major limitation to the applicability of Reinforcement Learning (RL) to many practical domains is the large number of samples required to learn an optimal policy. To address this problem and improve learning efficiency, we consider a linear hierarchy of abstraction layers of the Markov Decision Process (MDP) underlying the target domain. Each layer is an MDP representing a coarser model of the one immediately below in the hierarchy. In this work, we propose a novel form of Reward Shaping where the solution obtained at the abstract level is used to offer rewards to the more concrete MDP, in such a way that the abstract solution guides the learning in the more complex domain. In contrast with other works in Hierarchical RL, our technique has few requirements in the design of the abstract models and it is also tolerant to modeling errors, thus making the proposed approach practical. We formally analyze the relationship between the abstract models and the exploration heuristic induced in the lower-level domain. Moreover, we prove that the method guarantees optimal convergence and we demonstrate its effectiveness experimentally.

Exploiting Multiple Abstractions in Episodic RL via Reward Shaping

Most programmers make mistakes when writing code. Some of these mistakes are small and require few edits to the original program -- a class of errors recently termed last mile mistakes. These errors break the flow for experienced developers and can stump novice programmers. Existing automated repair techniques targeting this class of errors are language-specific and do not easily carry over to new languages. Transferring symbolic approaches requires substantial engineering and neural approaches require data and retraining. We introduce RING, a multilingual repair engine powered by a large language model trained on code (LLMC) such as Codex. Such a multilingual engine enables a flipped model for programming assistance, one where the programmer writes code and the AI assistance suggests fixes, compared to traditional code suggestion technology. Taking inspiration from the way programmers manually fix bugs, we show that a prompt-based strategy that conceptualizes repair as localization, transformation, and candidate ranking, can successfully repair programs in multiple languages with minimal effort. We present the first results for such a multilingual repair engine by evaluating on 6 different languages and comparing performance to language-specific repair engines. We show that RING can outperform language-specific repair engines for three of these languages.

Repair Is Nearly Generation: Multilingual Program Repair with LLMs

Domain generalisation (i.e. out-of-distribution generalisation)
is an open problem in machine learning, where the goal is
to train a model via one or more source domains, that will
generalise well to unknown target domains. While the topic is
attracting increasing interest, it has not been studied in detail
in the context of object detection. The established approaches
all operate under the covariate shift assumption, where the
conditional distributions are assumed to be approximately
equal across source domains. This is the first paper to address
domain generalisation in the context of object detection, with
a rigorous mathematical analysis of domain shift, without
the covariate shift assumption. "We focus on improving the
generalisation ability of object detection by proposing new
regularisation terms to address the domain shift that arises due
to both classification and bounding box regression. Also, we
include an additional consistency regularisation term to align
the local and global level predictions. The proposed approach
is evaluated using four object detection datasets which provide
domain metadata (GWHD, Cityscapes, BDD100K, Sim10K)
where it exhibits a consistent performance improvement over
the baselines.

Domain Generalised Faster R-CNN

In this paper, we identify symmetry property in adversarial scenario by viewing adversarial attack in a fine-grained manner. A newly designed metric called attack proportion, is thus proposed to count the proportion of the adversarial examples misclassified between classes. We observe that the distribution of attack proportion is unbalanced as each class shows vulnerability to particular classes. Further, some class pairs correlate strongly and have the same degree of attack proportion for each other. We call this intriguing phenomenon symmetry property. We empirically prove this phenomenon is widespread and then analyze the reason behind the existence of symmetry property. This explanation, to some extent, could be utilized to understand robust models, which also inspires us to strengthen adversarial defenses.

Towards Interpreting and Utilizing Symmetry Property in Adversarial Examples

We introduce equi-tuning, a novel fine-tuning method that transforms (potentially non-equivariant) pretrained models into group equivariant models while incurring minimum $L_2$ loss between the feature representations of the pretrained and the equivariant models. Large pretrained models can be equi-tuned for different groups to satisfy the needs of various downstream tasks. Equi-tuned models benefit from both group equivariance as an inductive bias and semantic priors from pretrained models. We provide applications of equi-tuning on three different tasks: image classification, compositional generalization in language, and fairness in natural language generation (NLG). We also provide a novel group-theoretic definition for fairness in NLG. The effectiveness of this definition is shown by testing it against a standard empirical method of fairness in NLG. We provide experimental results for equi-tuning using a variety of pretrained models: Alexnet, Resnet, VGG, and Densenet for image classification; RNNs, GRUs, and LSTMs for compositional generalization; and GPT2 for fairness in NLG. We test these models on benchmark datasets across all considered tasks to show the generality and effectiveness of the proposed method.

Equi-Tuning: Group Equivariant Fine-Tuning of Pretrained Models

We propose ordering-based approaches for learning the maximal ancestral graph (MAG) of a structural equation model (SEM) up to its Markov equivalence class (MEC) in the presence of unobserved variables. Existing ordering-based methods in the literature recover a graph through learning a causal order (c-order). We advocate for a novel order called removable order (r-order) as they are advantageous over c-orders for structure learning. This is because r-orders are the minimizers of an appropriately defined optimization problem that could be either solved exactly (using a reinforcement learning approach) or approximately (using a hill-climbing search). Moreover, the r-orders (unlike c-orders) are invariant among all the graphs in a MEC and include c-orders as a subset. Given that set of r-orders is often significantly larger than the set of c-orders, it is easier for the optimization problem to find an r-order instead of a c-order. We evaluate the performance and the scalability of our proposed approaches on both real-world and randomly generated networks.

Novel Ordering-based Approaches for Causal Structure Learning in the Presence of Unobserved Variables

  Herein, we propose a Spearman rank correlation-based screening procedure for ultrahigh-dimensional data with censored response cases. The proposed method is model-free without specifying any regression forms of predictors or response variables and is robust under the unknown monotone transformations of these response variable and predictors. The sure-screening and rank-consistency properties are established under some mild regularity conditions. Simulation studies demonstrate that the new screening method performs well in the presence
of a heavy-tailed distribution, strongly dependent predictors or outliers, and offers superior performance over the existing nonparametric screening procedures. In particular, the new screening method still works well when a response variable is observed under a high censoring rate. An illustrative example is provided.

Spearman Rank Correlation Screening for Ultrahigh-dimensional Censored Data

Determining kernels' sizes of a CNN model is a crucial and non-trivial design choice and has a significant impact on its performance. The majority of kernel size design methods rely on complex heuristic tricks or leverage neural architecture search that requires extreme computational resources. Thus, learning kernels' sizes jointly with the classifier's weights has been proposed as a workaround. However, the proposed methods cannot achieve satisfactory results or are inefficient for large-scale datasets. To fill in this gap, we design a novel efficient kernel size learning method in which a size predictor model learns to predict optimal kernels' sizes for a classifier given a desired number of parameters. It does so in collaboration with a hypernetwork that predicts the weights of the classifier's kernels - given kernel sizes predicted by the kernel predictor - to minimize the training objective and both models are trained end-to-end. Our method only needs a small fraction of training epochs of the original classifier to train these two models and find proper kernel sizes for it. Thus, it offers an efficient and effective solution for the kernel size learning problem. Our extensive experiments on MNIST, CIFAR-10, STL-10, and ImageNet-32 clearly demonstrates that our method can achieve the best training time vs. accuracy trade-off compared to previous kernel size learning methods and significantly outperform them on challenging datasets such as STL-10 and ImageNet-32.

Downloads

Next from AAAI 2023

Reconstructing an Epidemic Outbreak Using Steiner Connectivity

Similar lecture

Combining Curriculum Learning and Knowledge Distillation for Dialogue Generation