Many existing motion prediction approaches rely on symbolic perception outputs to generate agent trajectories, such as bounding boxes, road graph information and traffic lights. This symbolic representation is a high-level abstraction of the real world, which may render the motion prediction model vulnerable to perception errors (e.g., failures in detecting open-vocabulary obstacles) while missing salient information from the scene context (e.g., poor road conditions). An alternative paradigm is end-to-end learning from raw sensors. However, this approach suffers from the lack of interpretability and requires significantly more training resources. In this work, we propose tokenizing the visual world into a compact set of scene elements and then leveraging pre-trained image foundation models and LiDAR neural networks to encode all the scene elements in an open-vocabulary manner. The image foundation model enables our scene tokens to encode the general knowledge of the open world while the LiDAR neural network encodes geometry information. Our proposed representation can efficiently encode the multi-frame multi-modality observations with a few hundred tokens and is compatible with most transformer-based architectures. To evaluate our method, we have augmented Waymo Open Motion Dataset with camera embeddings. Experiments over Waymo Open Motion Dataset show that our approach leads to significant performance improvements over the state-of-the-art.

Motion forecasting has become an increasingly critical component of autonomous robotic systems. Onboard compute budgets typically limit the accuracy of real-time systems. In this work we propose methods of improving motion forecasting systems subject to limited compute budgets by combining model ensemble and distillation techniques. The use of ensembles of deep neural networks has been shown to improve generalization accuracy in many application domains. We first demonstrate significant performance gains by creating a large ensemble of optimized single models. We then develop a generalized framework to distill motion forecasting model ensembles into small student models which retain high performance with a fraction of the computing cost. For this study we focus on the task of motion forecasting using real world data from autonomous driving systems. We develop ensemble models that are very competitive on the Waymo Open Motion Dataset (WOMD) and Argoverse leaderboards. From these ensembles, we train distilled student models which have high performance at a fraction of the compute costs. These experiments demonstrate distillation from ensembles as an effective method for improving accuracy of predictive models for robotic systems with limited compute budgets.

Reliable forecasting of the future behavior of road agents is a critical component to safe planning in autonomous vehicles. Here, we represent continuous trajectories as sequences of discrete motion tokens and cast multi-agent motion prediction as a language modeling task over this domain. Our model, MotionLM, provides several advantages: First, it does not require anchors or explicit latent variable optimization to learn multimodal distributions. Instead, we leverage a single standard language modeling objective, maximizing the average log probability over sequence tokens. Second, our approach bypasses post-hoc interaction heuristics where individual agent trajectory generation is conducted prior to interactive scoring. Instead, MotionLM produces joint distributions over interactive agent futures in a single autoregressive decoding process. In addition, the model’s sequential factorization enables temporally causal conditional rollouts. The proposed approach establishes new state-of-the-art performance for multi-agent motion prediction on the Waymo Open Motion Dataset, ranking 1ston the interactive challenge leaderboard.

Motion forecasting for autonomous driving is a challenging task because complex driving scenarios involve a heterogeneous mix of static and dynamic inputs. It is an open problem how best to represent and fuse information about road geometry, lane connectivity, time-varying traffic light state, and history of a dynamic set of agents and their interactions into an effective encoding. To model this diverse set of input features, many approaches proposed to design an equally complex system with a diverse set of modality specific modules. This results in systems that are difficult to scale, extend, or tune in rigorous ways to trade off quality and efficiency. In this paper, we present Wayformer, a family of simple and homogeneous attention based architectures for motion forecasting. Wayformer offers a compact model description consisting of an attention based scene encoder and a decoder. In the scene encoder we study the choice of early, late and hierarchical fusion of input modalities. For each fusion type we explore strategies to trade off efficiency and quality via factorized attention or latent query attention. We show that early fusion, despite its simplicity, is not only modality agnostic but also achieves state-of-the-art results on both Waymo Open Motion Dataset (WOMD) and Argoverse leaderboards, demonstrating the effectiveness of our design philosophy.

There has been growing interest in parameter-efficient methods to apply pre-trained language models to downstream tasks. Building on the Prompt Tuning approach of Lester et al. (2021), which learns task-specific soft prompts to condition a frozen pre-trained model to perform different tasks, we propose a novel prompt-based transfer learning approach called SPoT: Soft Prompt Transfer. SPoT first learns a prompt on one or more source tasks and then uses it to initialize the prompt for a target task. We show that SPoT significantly boosts the performance of Prompt Tuning across many tasks. More remarkably, across all model sizes, SPoT matches or outperforms standard Model Tuning (which fine-tunes all model parameters) on the SuperGLUE benchmark, while using up to 27,000× fewer task-specific parameters. To understand where SPoT is most effective, we conduct a large-scale study on task transferability with 26 NLP tasks in 160 combinations, and demonstrate that many tasks can benefit each other via prompt transfer. Finally, we propose an efficient retrieval approach that interprets task prompts as task embeddings to identify similar tasks and predict the most transferable source tasks for a novel target task.

Behavior prediction models have proliferated in recent years, especially in the popular real-world robotics application of autonomous driving, where representing the distribution over possible futures of moving agents is essential for safe and comfortable motion planning. In these models, the choice of coordinate frames to represent inputs and outputs has crucial trade offs which broadly fall into one of two categories. Agent-centric models transform inputs and perform inference in agent-centric coordinates. These models are intrinsically invari-ant to translation and rotation between scene elements, are best-performing on public leaderboards, but scale quadratically with the number of agents and scene elements. Scene-centric models use a fixed coordinate system to process all agents. This gives them the advantage of sharing representations among all agents, offering efficient amortized inference computation which scales linearly with the number of agents. However, these models have to learn invariance to translation and rotation between scene elements, and typically underperform agent-centric models. In this work, we develop knowledge distillation techniques between probabilistic motion forecasting models, and apply these techniques to close the gap in performance between agent-centric and scene-centric models. This improves scene-centric model performance by 13.2% on the public Argoverse benchmark, 7.8% on Waymo Open Dataset and up to 9.4% on a large In-House dataset. These improved scene-centric models rank highly in public leaderboards and are up to 15 times more efficient than their agent-centric teacher counterparts in busy scenes.

Abstract Most widely used pre-trained language models operate on sequences of tokens corresponding to word or subword units. By comparison, token-free models that operate directly on raw text (bytes or characters) have many benefits: They can process text in any language out of the box, they are more robust to noise, and they minimize technical debt by removing complex and error-prone text preprocessing pipelines. Because byte or character sequences are longer than token sequences, past work on token-free models has often introduced new model architectures designed to amortize the cost of operating directly on raw text. In this paper, we show that a standard Transformer architecture can be used with minimal modifications to process byte sequences. We characterize the trade-offs in terms of parameter count, training FLOPs, and inference speed, and show that byte-level models are competitive with their token-level counterparts. We also demonstrate that byte-level models are significantly more robust to noise and perform better on tasks that are sensitive to spelling and pronunciation. As part of our contribution, we release a new set of pre-trained byte-level Transformer models based on the T5 architecture, as well as all code and data used in our experiments.1

In this work, we explore "prompt tuning," a simple yet effective mechanism for learning "soft prompts" to condition frozen language models to perform specific downstream tasks.Unlike the discrete text prompts used by GPT-3, soft prompts are learned through backpropagation and can be tuned to incorporate signals from any number of labeled examples.Our end-to-end learned approach outperforms GPT-3's few-shot learning by a large margin.More remarkably, through ablations on model size using T5, we show that prompt tuning becomes more competitive with scale: as models exceed billions of parameters, our method "closes the gap" and matches the strong performance of model tuning (where all model weights are tuned).This finding is especially relevant because large models are costly to share and serve and the ability to reuse one frozen model for multiple downstream tasks can ease this burden.Our method can be seen as a simplification of the recently proposed "prefix tuning" of Li and Liang (2021) and we provide a comparison to this and other similar approaches.Finally, we show that conditioning a frozen model with soft prompts confers benefits in robustness to domain transfer and enables efficient "prompt ensembling."We release code and model checkpoints to reproduce our experiments.1

Oshin Agarwal, Heming Ge, Siamak Shakeri, Rami Al-Rfou. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

We present LAReQA, a challenging new benchmark for language-agnostic answer retrieval from a multilingual candidate pool.Unlike previous cross-lingual tasks, LAReQA tests for "strong" cross-lingual alignment, requiring semantically related cross-language pairs to be closer in representation space than unrelated same-language pairs.Building on multilingual BERT (mBERT), we study different strategies for achieving strong alignment.We find that augmenting training data via machine translation is effective, and improves significantly over using mBERT out-of-the-box.Interestingly, the embedding baseline that performs the best on LAReQA falls short of competing baselines on zero-shot variants of our task that only target "weak" alignment.This finding underscores our claim that languageagnostic retrieval is a substantively new kind of cross-lingual evaluation.

We propose a new multilingual language model benchmark that is composed of 40+ languages spanning several scripts and linguistic families. With around 40 billion characters, we hope this new resource will accelerate the research of multilingual modeling. We train monolingual causal language models using a state-of-the-art model (Transformer-XL) establishing baselines for many languages. We also introduce the task of multilingual causal language modeling where we train our model on the combined text of 40+ languages from Wikipedia with different vocabulary sizes and evaluate on the languages individually. We released the cleaned-up text of 40+ Wikipedia language editions, the corresponding trained monolingual language models, and several multilingual language models with different fixed vocabulary sizes.

LSTMs and other RNN variants have shown strong performance on character-level language modeling. These models are typically trained using truncated backpropagation through time, and it is common to assume that their success stems from their ability to remember long-term contexts. In this paper, we show that a deep (64-layer) transformer model (Vaswani et al. 2017) with fixed context outperforms RNN variants by a large margin, achieving state of the art on two popular benchmarks: 1.13 bits per character on text8 and 1.06 on enwik8. To get good results at this depth, we show that it is important to add auxiliary losses, both at intermediate network layers and intermediate sequence positions.

Can neural networks learn to compare graphs without feature engineering? In this paper, we show that it is possible to learn representations for graph similarity with neither domain knowledge nor supervision (i.e. feature engineering or labeled graphs). We propose Deep Divergence Graph Kernels, an unsupervised method for learning representations over graphs that encodes a relaxed notion of graph isomorphism. Our method consists of three parts. First, we learn an encoder for each anchor graph to capture its structure. Second, for each pair of graphs, we train a cross-graph attention network which uses the node representations of an anchor graph to reconstruct another graph. This approach, which we call isomorphism attention, captures how well the representations of one graph can encode another. We use the attention-augmented encoder's predictions to define a divergence score for each pair of graphs. Finally, we construct an embedding space for all graphs using these pair-wise divergence scores.

Graph embedding methods represent nodes in a continuous vector space, preserving different types of relational information from the graph. There are many hyper-parameters to these methods (e.g. the length of a random walk) which have to be manually tuned for every graph. In this paper, we replace previously fixed hyper-parameters with trainable ones that we automatically learn via backpropagation. In particular, we propose a novel attention model on the power series of the transition matrix, which guides the random walk to optimize an upstream objective. Unlike previous approaches to attention models, the method that we propose utilizes attention parameters exclusively on the data itself (e.g. on the random walk), and are not used by the model for inference. We experiment on link prediction tasks, as we aim to produce embeddings that best-preserve the graph structure, generalizing to unseen information. We improve state-of-the-art results on a comprehensive suite of real-world graph datasets including social, collaboration, and biological networks, where we observe that our graph attention model can reduce the error by up to 20\%-40\%. We show that our automatically-learned attention parameters can vary significantly per graph, and correspond to the optimal choice of hyper-parameter if we manually tune existing methods.

We propose a new method for embedding graphs while preserving directed edge information. Learning such continuous-space vector representations (or embeddings) of nodes in a graph is an important first step for using network information (from social networks, user-item graphs, knowledge bases, etc.) in many machine learning tasks. Unlike previous work, we (1) explicitly model an edge as a function of node embeddings, and we (2) propose a novel objective, the graph likelihood, which contrasts information from sampled random walks with non-existent edges. Individually, both of these contributions improve the learned representations, especially when there are memory constraints on the total size of the embeddings. When combined, our contributions enable us to significantly improve the state-of-the-art by learning more concise representations that better preserve the graph structure. We evaluate our method on a variety of link-prediction task including social networks, collaboration networks, and protein interactions, showing that our proposed method learn representations with error reductions of up to 76% and 55%, on directed and undirected graphs. In addition, we show that the representations learned by our method are quite space efficient, producing embeddings which have higher structure-preserving accuracy but are 10 times smaller.

The increasing diversity of languages used on the web introduces a new level of complexity to Information Retrieval (IR) systems. We can no longer assume that textual content is written in one language or even the same language family. In this paper, we demonstrate how to build massive multilingual annotators with minimal human expertise and intervention. We describe a system that builds Named Entity Recognition (NER) annotators for 40 major languages using Wikipedia and Freebase. Our approach does not require NER human annotated datasets or language specific resources like treebanks, parallel corpora, and orthographic rules. The novelty of approach lies therein - using only language agnostic techniques, while achieving competitive performance. Our method learns distributed word representations (word embeddings) which encode semantic and syntactic features of words in each language. Then, we automatically generate datasets from Wikipedia link structure and Freebase attributes. Finally, we apply two preprocessing stages (oversampling and exact surface form matching) which do not require any linguistic expertise. Our evaluation is two fold: First, we demonstrate the system performance on human annotated datasets. Second, for languages where no gold-standard benchmarks are available, we propose a new method, distant evaluation, based on statistical machine translation.

We propose a new computational approach for tracking and detecting statistically significant linguistic shifts in the meaning and usage of words. Such linguistic shifts are especially prevalent on the Internet, where the rapid exchange of ideas can quickly change a word's meaning. Our meta-analysis approach constructs property time series of word usage, and then uses statistically sound change point detection algorithms to identify significant linguistic shifts. We consider and analyze three approaches of increasing complexity to generate such linguistic property time series, the culmination of which uses distributional characteristics inferred from word co-occurrences. Using recently proposed deep neural language models, we first train vector representations of words for each time period. Second, we warp the vector spaces into one unified coordinate system. Finally, we construct a distance-based distributional time series for each word to track its linguistic displacement over time.

We present DeepWalk, a novel approach for learning latent representations of vertices in a network. These latent representations encode social relations in a continuous vector space, which is easily exploited by statistical models. DeepWalk generalizes recent advancements in language modeling and unsupervised feature learning (or deep learning) from sequences of words to graphs.