An official website of the United States government
Knowledge Distillation (KD) (Hinton et al., 2015) is one of the most effective approaches for deploying large-scale pre-trained language models in low-latency environments by transferring the knowledge contained in the largescale models to smaller student models. Previous KD approaches use the soft labels and intermediate activations generated by the teacher to transfer knowledge to the student model parameters alone. In this paper, we show that having access to non-parametric memory in the form of a knowledge base with the teacher’s soft labels and predictions can further enhance student capacity and improve generalization. To enable the student to retrieve from the knowledge base effectively, we propose a new Retrieval-augmented KD framework with a loss function that aligns the relational knowledge in teacher and student embedding spaces. We show through extensive experiments that our retrieval mechanism can achieve state-of-the-art performance for taskspecific knowledge distillation on the GLUE benchmark (Wang et al., 2018a).
more »
« less
Specialized Embedding Approximation for Edge Intelligence: A Case Study in Urban Sound Classification
Embedding models that encode semantic information into low-dimensional vector representations are useful in various machine learning tasks with limited training data. However, these models are typically too large to support inference in small edge devices, which motivates training of smaller yet comparably predictive student embedding models through knowledge distillation (KD). While knowledge distillation traditionally uses the teacher’s original training dataset to train the student, we hypothesize that using a dataset similar to the student’s target domain allows for better compression and training efficiency for the said domain, at the cost of reduced generality across other (non-pertinent) domains. Hence, we introduce Specialized Embedding Approximation (SEA) to train a student featurizer to approximate the teacher’s embedding manifold for a given target domain. We demonstrate the feasibility of SEA in the context of acoustic event classification for urban noise monitoring and show that leveraging a dataset related to this target domain not only improves the baseline performance of the original embedding model but also yields competitive students with >1 order of magnitude lesser storage and activation memory. We further investigate the impact of using random and informed sampling techniques for dimensionality reduction in SEA.
more »
« less
- Award ID(s):
- 2026704
- PAR ID:
- 10295446
- Date Published:
- Journal Name:
- ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
- Page Range / eLocation ID:
- 8378 to 8382
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Knowledge distillation deploys complex machine learning models in resource-constrained environments by training a smaller student model to emulate internal representations of a complex teacher model. However, the teacher’s representations can also encode nuisance or additional information not relevant to the downstream task. Distilling such irrelevant information can actually impede the performance of a capacity-limited student model. This observation motivates our primary question: What are the information-theoretic limits of knowledge distillation? To this end, we leverage Partial Information Decomposition to quantify and explain the transferred knowledge and knowledge left to distill for a downstream task. We theoretically demonstrate that the task-relevant transferred knowledge is succinctly captured by the measure of redundant information about the task between the teacher and student. We propose a novel multi-level optimization to incorporate redundant information as a regularizer, leading to our framework of Redundant Information Distillation (RID). RID leads to more resilient and effective distillation under nuisance teachers as it succinctly quantifies task-relevant knowledge rather than simply aligning student and teacher representations.more » « less
-
Knowledge distillation aims at reducing model size without compromising much performance. Recent work has applied it to large vision-language (VL) Transformers, and has shown that attention maps in the multi-head attention modules of vision-language Transformers contain extensive intra-modal and cross-modal co-reference relations to be distilled. The standard approach is to apply a one-to-one attention map distillation loss, i.e. the Teacher’s first attention head instructs the Student’s first head, the second teaches the second, and so forth, but this only works when the numbers of attention heads in the Teacher and Student are the same. To remove this constraint, we propose a new Attention Map Alignment Distillation (AMAD) method for Transformers with multi-head attention, which works for a Teacher and a Student with different numbers of attention heads. Specifically, we soft-align different heads in Teacher and Student attention maps using a cosine similarity weighting. The Teacher head contributes more to the Student heads for which it has a higher similarity weight. Each Teacher head contributes to all the Student heads by minimizing the divergence between the attention activation distributions for the soft-aligned heads. No head is left behind. This distillation approach operates like cross-attention. We experiment on distilling VL-T5 and BLIP, and apply AMAD loss on their T5, BERT, and ViT sub-modules. We show, under vision-language setting, that AMAD outperforms conventional distillation methods on VQA-2.0, COCO captioning, and Multi30K translation datasets. We further show that even without VL pre-training, the distilled VL-T5 models outperform corresponding VL pre-trained VL-T5 models that are further fine-tuned by ground-truth signals, and that fine-tuning distillation can also compensate to some degree for the absence of VL pre-training for BLIP models.more » « less
-
We propose a novel knowledge distillation (KD) method to selectively instill teacher knowledge into a student model motivated by situations where the student’s capacity is significantly smaller than that of the teachers. In vanilla KD, the teacher primarily sets a predictive target for the student to follow, and we posit that this target is overly optimistic due to the student’s lack of capacity. We develop a novel scaffolding scheme where the teacher, in addition to setting a predictive target, also scaffolds the student’s prediction by censoring hard-to-learn examples. The student model utilizes the same information as the teacher’s soft-max predictions as inputs, and in this sense, our proposal can be viewed as a natural variant of vanilla KD. We show on synthetic examples that censoring hard-examples leads to smoothening the student’s loss landscape so that the student encounters fewer local minima. As a result, it has good generalization properties. Against vanilla KD, we achieve improved performance and are comparable to more intrusive techniques that leverage feature matching on benchmark datasets.more » « less
-
Recent approaches have shown promises distilling diffusion models into efficient one-step generators. Among them, Distribution Matching Distillation (DMD) produces one-step generators that match their teacher in distribution, without enforcing a one-to-one correspondence with the sampling trajectories of their teachers. However, to ensure stable training, DMD requires an additional regression loss computed using a large set of noise-image pairs generated by the teacher with many steps of a deterministic sampler. This is costly for large-scale text-to-image synthesis and limits the student's quality, tying it too closely to the teacher's original sampling paths. We introduce DMD2, a set of techniques that lift this limitation and improve DMD training. First, we eliminate the regression loss and the need for expensive dataset construction. We show that the resulting instability is due to the fake critic not estimating the distribution of generated samples accurately and propose a two time-scale update rule as a remedy. Second, we integrate a GAN loss into the distillation procedure, discriminating between generated samples and real images. This lets us train the student model on real data, mitigating the imperfect real score estimation from the teacher model, and enhancing quality. Lastly, we modify the training procedure to enable multi-step sampling. We identify and address the training-inference input mismatch problem in this setting, by simulating inference-time generator samples during training time. Taken together, our improvements set new benchmarks in one-step image generation, with FID scores of 1.28 on ImageNet-64x64 and 8.35 on zero-shot COCO 2014, surpassing the original teacher despite a 500X reduction in inference cost. Further, we show our approach can generate megapixel images by distilling SDXL, demonstrating exceptional visual quality among few-step methods.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1109/ICASSP39728.2021.9414287
