Best of Both Worlds: Transferring Knowledge from Discriminative Learning to a Generative Visual Dialog Model

By Jiasen Lu, Anitha Kannan, Jianwei Yang, Devi Parikh, Dhruv Batra. 2017

139 citations recorded by the source corpus.

We present a novel training framework for neural sequence models, particularly for grounded dialog generation. The standard training paradigm for these models is maximum likelihood estimation (MLE), or minimizing the cross-entropy of the human responses. Across a variety of domains, a recurring problem with MLE trained generative neural dialog models (G) is that they tend to produce 'safe' and generic responses ("I don't know", "I can't tell"). In contrast, discriminative dialog models (D) that are trained to rank a list of candidate human responses outperform their generative counterparts; in terms of automatic metrics, diversity, and informativeness of the responses. However, D is not useful in practice since it cannot be deployed to have real conversations with users. Our work aims to achieve the best of both worlds -- the practical usefulness of G and the strong performance of D -- via knowledge transfer from D to G. Our primary contribution is an end-to-end trainable generative visual dialog model, where G receives gradients from D as a perceptual (not adversarial) loss of the sequence sampled from G. We leverage the recently proposed Gumbel-Softmax (GS) approximation to the discrete distribution -- specifically, an RNN augmented with a sequence of GS samplers, coupled with the straight-through gradient estimator to enable end-to-end differentiability. We also introduce a stronger encoder for visual dialog, and employ a self-attention mechanism for answer encoding along with a metric learning loss to aid D in better capturing semantic similarities in answer responses. Overall, our proposed model outperforms state-of-the-art on the VisDial dataset by a significant margin (2.67% on recall@10). The source code can be downloaded from this https URL

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Multimodal Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Multimodal Learning

Citations in the corpus

Large-scale Pretraining for Visual Dialog: A Simple State-of-the-Art Baseline (2019, background): “Recent years have seen incredible progress in Visual Dialog [1, 5–25], spurred in part by the initial efforts of Das et al.”
Multi-step Reasoning via Recurrent Dual Attention for Visual Dialog (2019, methodology): “Impact of image features Since VGG-Net [42] was widely used in previous visual dialog models, we ﬁrst compare the performance of using VGG-Net and the bottom-up-attention features based on HCIAE [32].”
Two Can Play This Game: Visual Dialog with Discriminative Question Generation and Answering (2018, methodology): “[23] use an attention based mechanism to combine all previous rounds of history to get a single representation.”
DialGraph: Sparse Graph Learning Networks for Visual Dialog (2020, methodology): “Typically, most of the previous research has focused on extracting the rich question-relevant representations from the given image and dialog history, while implicitly finding their relationships [8,28,42,13,9,38].”
Learn, Imagine and Create: Text-to-Image Generation from Prior Knowledge (2019, background): “They also boost deep model performance by guiding the generators to focus on the relevant information [43, 42, 40, 25, 37, 13, 3, 26, 12, 14, 32].”
Exemplar Encoder-Decoder for Neural Conversation Generation (2018, background): “Recently a framework has also been proposed that uses a discriminative dialog network that ranks the candidate responses received from a response generator network and trains both the networks in an end to end manner (Lu et al., 2017a).”
Asking the Difficult Questions: Goal-Oriented Visual Question Generation via Intermediate Rewards (2017, methodology): “Reinforcement Learning for V2L Reinforcement learning [14, 27] has been adopted in several vision to language (V2L) problems, including image captioning [17, 23, 24], VQA [2, 12, 33], and aforementioned visual dialogue sys-tem [8, 18] etc .”
Two Causal Principles for Improving Visual Dialog (2019, methodology): “We report the performance of the following base models, including LF [9], HCIAE [21], CoAtt [41] and RvA [26]: LF [9].”
Are You Talking to Me? Reasoned Visual Dialog Generation Through Adversarial Learning (2017, methodology): “MRR R@1 R@5 R@10 Mean Answer Prior [5] 0.3735 23.55 48.52 53.23 26.50 NN [5] 0.4274 33.13 50.83 58.69 19.62 LF [5] 0.5199 41.83 61.78 67.59 17.07 HRE [5] 0.5237 42.29 62.18 67.92 17.07 HREA [5] 0.5242 42.28 62.33 68.17 16.79 MN [5] 0.5259 42.29 62.85 68.88 17.06 HCIAE [18] 0.5386 44.06 63.55 69.”
Large-Scale Answerer in Questioner's Mind for Visual Dialog Question Generation (2019, methodology): “Therefore, the VQA task in the Visual Dialog dataset is much studied than the GuessWhat dataset (Lu et al., 2017; Seo et al., 2017b).”
Making History Matter: Gold-Critic Sequence Training for Visual Dialog (2019, methodology): “Thanks to the end-to-end deep neural networks in their respective sub-tasks, state-of-the-art visual dialog systems can be built by assembling them into a codec framework [31, 7, 21].”
Diversifying Dialogue Generation with Non-Conversational Text (2020, background): “The scores for diversity and relevance generally correlate well with the automatic evaluations. approaches in its capability to properly make use of the non-conversational corpus.”
Factor Graph Attention (2019, background): “[32, 33] combined a generative and discriminative model to choose generated answers, and also proposed history attention conditioned on the image using hierarchical co-attention developed for visual question answering.”
Image-Question-Answer Synergistic Network for Visual Dialog (2019, methodology): “Discriminative models tend to calculate the similarity between the latent output of the encoder and the embedding of candidate answers [4, 18, 12, 29], with the correct answer expected to have the highest score.”
Learning to Ground Visual Objects for Visual Dialog (2021, methodology): “, 2017), HCIAE (Lu et al., 2017), CorefNMN (Kottur et al.”
Semantic bottleneck for computer vision tasks (2018, background): “Another way to enrich the caption is to generate a set of questions/answers such as proposed in the Visual Dialog framework [1,13,14].”
Transfer Learning in Deep Reinforcement Learning: A Survey (2020, methodology): “Examples include learning from expert demonstrations for Spoken Dialogue Systems [154], VQA [152]; or reward shaping for Sequence Generation [155], Spoken Dialogue Systems [80], QA [81], [156], and Image Caption [153], or transferring policies for Structured Prediction [157] and VQA [158].”
Discriminability Objective for Training Descriptive Captions (2018, methodology): “In [42], more meaningful visual dialog responses are generated by distilling knowledge from a discriminative model trained to rank different dialog responses given the previous dialog context.”