Víctor Campos

Steven Kapturowski, Víctor Campos, Ray Jiang et al.

The task of building general agents that perform well over a wide range of tasks has been an important goal in reinforcement learning since its inception. The problem has been subject of research of a large body of work, with performance frequently measured by observing scores over the wide range of environments contained in the Atari 57 benchmark. Agent57 was the first agent to surpass the human benchmark on all 57 games, but this came at the cost of poor data-efficiency, requiring nearly 80 billion frames of experience to achieve. Taking Agent57 as a starting point, we employ a diverse set of strategies to achieve a 200-fold reduction of experience needed to out perform the human baseline. We investigate a range of instabilities and bottlenecks we encountered while reducing the data regime, and propose effective solutions to build a more robust and efficient agent. We also demonstrate competitive performance with high-performing methods such as Muesli and MuZero. The four key components to our approach are (1) an approximate trust region method which enables stable bootstrapping from the online network, (2) a normalisation scheme for the loss and priorities which improves robustness when learning a set of value functions with a wide range of scales, (3) an improved architecture employing techniques from NFNets in order to leverage deeper networks without the need for normalization layers, and (4) a policy distillation method which serves to smooth out the instantaneous greedy policy overtime.

Gemini Team, Petko Georgiev, Ving Ian Lei et al. · deepmind, mila

In this report, we introduce the Gemini 1.5 family of models, representing the next generation of highly compute-efficient multimodal models capable of recalling and reasoning over fine-grained information from millions of tokens of context, including multiple long documents and hours of video and audio. The family includes two new models: (1) an updated Gemini 1.5 Pro, which exceeds the February version on the great majority of capabilities and benchmarks; (2) Gemini 1.5 Flash, a more lightweight variant designed for efficiency with minimal regression in quality. Gemini 1.5 models achieve near-perfect recall on long-context retrieval tasks across modalities, improve the state-of-the-art in long-document QA, long-video QA and long-context ASR, and match or surpass Gemini 1.0 Ultra's state-of-the-art performance across a broad set of benchmarks. Studying the limits of Gemini 1.5's long-context ability, we find continued improvement in next-token prediction and near-perfect retrieval (>99%) up to at least 10M tokens, a generational leap over existing models such as Claude 3.0 (200k) and GPT-4 Turbo (128k). Finally, we highlight real-world use cases, such as Gemini 1.5 collaborating with professionals on completing their tasks achieving 26 to 75% time savings across 10 different job categories, as well as surprising new capabilities of large language models at the frontier; when given a grammar manual for Kalamang, a language with fewer than 200 speakers worldwide, the model learns to translate English to Kalamang at a similar level to a person who learned from the same content.

Víctor Campos, Alexander Trott, Caiming Xiong et al.

Acquiring abilities in the absence of a task-oriented reward function is at the frontier of reinforcement learning research. This problem has been studied through the lens of empowerment, which draws a connection between option discovery and information theory. Information-theoretic skill discovery methods have garnered much interest from the community, but little research has been conducted in understanding their limitations. Through theoretical analysis and empirical evidence, we show that existing algorithms suffer from a common limitation -- they discover options that provide a poor coverage of the state space. In light of this, we propose 'Explore, Discover and Learn' (EDL), an alternative approach to information-theoretic skill discovery. Crucially, EDL optimizes the same information-theoretic objective derived from the empowerment literature, but addresses the optimization problem using different machinery. We perform an extensive evaluation of skill discovery methods on controlled environments and show that EDL offers significant advantages, such as overcoming the coverage problem, reducing the dependence of learned skills on the initial state, and allowing the user to define a prior over which behaviors should be learned. Code is publicly available at https://github.com/victorcampos7/edl.

Víctor Campos, Pablo Sprechmann, Steven Hansen et al.

Designing agents that acquire knowledge autonomously and use it to solve new tasks efficiently is an important challenge in reinforcement learning. Knowledge acquired during an unsupervised pre-training phase is often transferred by fine-tuning neural network weights once rewards are exposed, as is common practice in supervised domains. Given the nature of the reinforcement learning problem, we argue that standard fine-tuning strategies alone are not enough for efficient transfer in challenging domains. We introduce Behavior Transfer (BT), a technique that leverages pre-trained policies for exploration and that is complementary to transferring neural network weights. Our experiments show that, when combined with large-scale pre-training in the absence of rewards, existing intrinsic motivation objectives can lead to the emergence of complex behaviors. These pre-trained policies can then be leveraged by BT to discover better solutions than without pre-training, and combining BT with standard fine-tuning strategies results in additional benefits. The largest gains are generally observed in domains requiring structured exploration, including settings where the behavior of the pre-trained policies is misaligned with the downstream task.

Devansh Arpit, Victor Campos, Yoshua Bengio

Residual networks (ResNet) and weight normalization play an important role in various deep learning applications. However, parameter initialization strategies have not been studied previously for weight normalized networks and, in practice, initialization methods designed for un-normalized networks are used as a proxy. Similarly, initialization for ResNets have also been studied for un-normalized networks and often under simplified settings ignoring the shortcut connection. To address these issues, we propose a novel parameter initialization strategy that avoids explosion/vanishment of information across layers for weight normalized networks with and without residual connections. The proposed strategy is based on a theoretical analysis using mean field approximation. We run over 2,500 experiments and evaluate our proposal on image datasets showing that the proposed initialization outperforms existing initialization methods in terms of generalization performance, robustness to hyper-parameter values and variance between seeds, especially when networks get deeper in which case existing methods fail to even start training. Finally, we show that using our initialization in conjunction with learning rate warmup is able to reduce the gap between the performance of weight normalized and batch normalized networks.

Víctor Campos, Xavier Giro-i-Nieto, Jordi Torres

Evolution Strategies (ES) emerged as a scalable alternative to popular Reinforcement Learning (RL) techniques, providing an almost perfect speedup when distributed across hundreds of CPU cores thanks to a reduced communication overhead. Despite providing large improvements in wall-clock time, ES is data inefficient when compared to competing RL methods. One of the main causes of such inefficiency is the collection of large batches of experience, which are discarded after each policy update. In this work, we study how to perform more than one update per batch of experience by means of Importance Sampling while preserving the scalability of the original method. The proposed method, Importance Weighted Evolution Strategies (IW-ES), shows promising results and is a first step towards designing efficient ES algorithms.

Daniel Fojo, Víctor Campos, Xavier Giro-i-Nieto

Adaptive Computation Time for Recurrent Neural Networks (ACT) is one of the most promising architectures for variable computation. ACT adapts to the input sequence by being able to look at each sample more than once, and learn how many times it should do it. In this paper, we compare ACT to Repeat-RNN, a novel architecture based on repeating each sample a fixed number of times. We found surprising results, where Repeat-RNN performs as good as ACT in the selected tasks. Source code in TensorFlow and PyTorch is publicly available at https://imatge-upc.github.io/danifojo-2018-repeatrnn/

Victor Campos, Brendan Jou, Xavier Giro-i-Nieto et al.

Recurrent Neural Networks (RNNs) continue to show outstanding performance in sequence modeling tasks. However, training RNNs on long sequences often face challenges like slow inference, vanishing gradients and difficulty in capturing long term dependencies. In backpropagation through time settings, these issues are tightly coupled with the large, sequential computational graph resulting from unfolding the RNN in time. We introduce the Skip RNN model which extends existing RNN models by learning to skip state updates and shortens the effective size of the computational graph. This model can also be encouraged to perform fewer state updates through a budget constraint. We evaluate the proposed model on various tasks and show how it can reduce the number of required RNN updates while preserving, and sometimes even improving, the performance of the baseline RNN models. Source code is publicly available at https://imatge-upc.github.io/skiprnn-2017-telecombcn/ .

Delia Fernandez, Alejandro Woodward, Victor Campos et al.

The increasing availability of affect-rich multimedia resources has bolstered interest in understanding sentiment and emotions in and from visual content. Adjective-noun pairs (ANP) are a popular mid-level semantic construct for capturing affect via visually detectable concepts such as "cute dog" or "beautiful landscape". Current state-of-the-art methods approach ANP prediction by considering each of these compound concepts as individual tokens, ignoring the underlying relationships in ANPs. This work aims at disentangling the contributions of the `adjectives' and `nouns' in the visual prediction of ANPs. Two specialised classifiers, one trained for detecting adjectives and another for nouns, are fused to predict 553 different ANPs. The resulting ANP prediction model is more interpretable as it allows us to study contributions of the adjective and noun components. Source code and models are available at https://imatge-upc.github.io/affective-2017-musa2/ .

Xunyu Lin, Victor Campos, Xavier Giro-i-Nieto et al.

This paper introduces an unsupervised framework to extract semantically rich features for video representation. Inspired by how the human visual system groups objects based on motion cues, we propose a deep convolutional neural network that disentangles motion, foreground and background information. The proposed architecture consists of a 3D convolutional feature encoder for blocks of 16 frames, which is trained for reconstruction tasks over the first and last frames of the sequence. A preliminary supervised experiment was conducted to verify the feasibility of proposed method by training the model with a fraction of videos from the UCF-101 dataset taking as ground truth the bounding boxes around the activity regions. Qualitative results indicate that the network can successfully segment foreground and background in videos as well as update the foreground appearance based on disentangled motion features. The benefits of these learned features are shown in a discriminative classification task, where initializing the network with the proposed pretraining method outperforms both random initialization and autoencoder pretraining. Our model and source code are publicly available at https://imatge-upc.github.io/unsupervised-2017-cvprw/ .

Victor Campos, Brendan Jou, Xavier Giro-i-Nieto

Visual multimedia have become an inseparable part of our digital social lives, and they often capture moments tied with deep affections. Automated visual sentiment analysis tools can provide a means of extracting the rich feelings and latent dispositions embedded in these media. In this work, we explore how Convolutional Neural Networks (CNNs), a now de facto computational machine learning tool particularly in the area of Computer Vision, can be specifically applied to the task of visual sentiment prediction. We accomplish this through fine-tuning experiments using a state-of-the-art CNN and via rigorous architecture analysis, we present several modifications that lead to accuracy improvements over prior art on a dataset of images from a popular social media platform. We additionally present visualizations of local patterns that the network learned to associate with image sentiment for insight into how visual positivity (or negativity) is perceived by the model.

Victor Campos, Amaia Salvador, Brendan Jou et al.

Visual media are powerful means of expressing emotions and sentiments. The constant generation of new content in social networks highlights the need of automated visual sentiment analysis tools. While Convolutional Neural Networks (CNNs) have established a new state-of-the-art in several vision problems, their application to the task of sentiment analysis is mostly unexplored and there are few studies regarding how to design CNNs for this purpose. In this work, we study the suitability of fine-tuning a CNN for visual sentiment prediction as well as explore performance boosting techniques within this deep learning setting. Finally, we provide a deep-dive analysis into a benchmark, state-of-the-art network architecture to gain insight about how to design patterns for CNNs on the task of visual sentiment prediction.