For fine-grained generation and recognition tasks such as minimally-supervised text-to-speech (TTS), voice conversion (VC), and automatic speech recognition (ASR), the intermediate representations extracted from speech should serve as a "bridge" between text and acoustic information, containing information from both modalities. The semantic content is emphasized, while the paralinguistic information such as speaker identity and acoustic details should be de-emphasized. However, existing methods for extracting fine-grained intermediate representations from speech suffer from issues of excessive redundancy and dimension explosion. Contrastive learning is a good method for modeling intermediate representations from two modalities. However, existing contrastive learning methods in the audio field focus on extracting global descriptive information for downstream audio classification tasks, making them unsuitable for TTS, VC, and ASR tasks. To address these issues, we propose a method named "Contrastive Token-Acoustic Pretraining (CTAP)", which uses two encoders to bring phoneme and speech into a joint multimodal space, learning how to connect phoneme and speech at the frame level. The CTAP model is trained on 210k speech and phoneme pairs, achieving minimally-supervised TTS, VC, and ASR. The proposed CTAP method offers a promising solution for fine-grained generation and recognition downstream tasks in speech processing. We provide a website with audio samples.

This report introduces a new family of multimodal models, Gemini, that exhibit remarkable capabilities across image, audio, video, and text understanding. The Gemini family consists of Ultra, Pro, and Nano sizes, suitable for applications ranging from complex reasoning tasks to on-device memory-constrained use-cases. Evaluation on a broad range of benchmarks shows that our most-capable Gemini Ultra model advances the state of the art in 30 of 32 of these benchmarks - notably being the first model to achieve human-expert performance on the well-studied exam benchmark MMLU, and improving the state of the art in every one of the 20 multimodal benchmarks we examined. We believe that the new capabilities of Gemini models in cross-modal reasoning and language understanding will enable a wide variety of use cases and we discuss our approach toward deploying them responsibly to users.

Recently, there has been a growing interest in text-to-speech (TTS) methods that can be trained with minimal supervision by combining two types of discrete speech representations and using two sequence-to-sequence tasks to decouple TTS. However, existing methods suffer from three problems: the high dimensionality and waveform distortion of discrete speech representations, the prosodic averaging problem caused by the duration prediction model in non-autoregressive frameworks, and the information redundancy and dimension explosion problems of existing semantic encoding methods. To address these problems, three progressive methods are proposed. First, we propose Diff-LM-Speech, an autoregressive structure consisting of a language model and diffusion models, which models the semantic embedding into the mel-spectrogram based on a diffusion model to achieve higher audio quality. We also introduce a prompt encoder structure based on a variational autoencoder and a prosody bottleneck to improve prompt representation ability. Second, we propose Tetra-Diff-Speech, a non-autoregressive structure consisting of four diffusion model-based modules that design a duration diffusion model to achieve diverse prosodic expressions. Finally, we propose Tri-Diff-Speech, a non-autoregressive structure consisting of three diffusion model-based modules that verify the non-necessity of existing semantic encoding models and achieve the best results. Experimental results show that our proposed methods outperform baseline methods. We provide a website with audio samples.

With the proliferation of IoT devices, researchers have developed a variety of IoT device identification methods with the assistance of machine learning. Nevertheless, the security of these identification methods mostly depends on collected training data. In this research, we propose a novel attack strategy named IoTGAN to manipulate an IoT device's traffic such that it can evade machine learning based IoT device identification. In the development of IoTGAN, we have two major technical challenges: (i) How to obtain the discriminative model in a black-box setting, and (ii) How to add perturbations to IoT traffic through the manipulative model, so as to evade the identification while not influencing the functionality of IoT devices. To address these challenges, a neural network based substitute model is used to fit the target model in black-box settings, it works as a discriminative model in IoTGAN. A manipulative model is trained to add adversarial perturbations into the IoT device's traffic to evade the substitute model. Experimental results show that IoTGAN can successfully achieve the attack goals. We also develop efficient countermeasures to protect machine learning based IoT device identification from been undermined by IoTGAN.

Overparameterized large-scale language models have impressive generalization performance of in-context few-shot learning. However, most language models allocate the same amount of parameters or computation to each token, disregarding the complexity or importance of the input data. We argue that in language model pretraining, a variable amount of computation should be assigned to different tokens, and this can be efficiently achieved via a simple routing mechanism. Different from conventional early stopping techniques where tokens can early exit at only early layers, we propose a more general method that dynamically skips the execution of a layer (or module) for any input token with a binary router. In our extensive evaluation across 24 NLP tasks, we demonstrate that the proposed method can significantly improve the 1-shot performance compared to other competitive baselines only at mild extra cost for inference.

Supervised learning algorithms based on Convolutional Neural Networks have become the benchmark for medical image segmentation tasks, but their effectiveness heavily relies on a large amount of labeled data. However, annotating medical image datasets is a laborious and time-consuming process. Inspired by semi-supervised algorithms that use both labeled and unlabeled data for training, we propose the PLGDF framework, which builds upon the mean teacher network for segmenting medical images with less annotation. We propose a novel pseudo-label utilization scheme, which combines labeled and unlabeled data to augment the dataset effectively. Additionally, we enforce the consistency between different scales in the decoder module of the segmentation network and propose a loss function suitable for evaluating the consistency. Moreover, we incorporate a sharpening operation on the predicted results, further enhancing the accuracy of the segmentation. Extensive experiments on three publicly available datasets demonstrate that the PLGDF framework can largely improve performance by incorporating the unlabeled data. Meanwhile, our framework yields superior performance compared to six state-of-the-art semi-supervised learning methods. The codes of this study are available at https://github.com/ortonwang/PLGDF.

ChatGPT has shown its great power in text processing, including its reasoning ability from text reading. However, there has not been any direct comparison between human readers and ChatGPT in reasoning ability related to text reading. This study was undertaken to investigate how ChatGPTs (i.e., ChatGPT and ChatGPT Plus) and Chinese senior school students as ESL learners exhibited their reasoning ability from English narrative texts. Additionally, we compared the two ChatGPTs in the reasoning performances when commands were updated elaborately. The whole study was composed of three reasoning tests: Test 1 for commonsense inference, Test 2 for emotional inference, and Test 3 for causal inference. The results showed that in Test 1, the students outdid the two ChatGPT versions in local-culture-related inferences but performed worse than the chatbots in daily-life inferences. In Test 2, ChatGPT Plus excelled whereas ChatGPT lagged behind in accuracy. In association with both accuracy and frequency of correct responses, the students were inferior to the two chatbots. Compared with ChatGPTs' better performance in positive emotions, the students showed their superiority in inferring negative emotions. In Test 3, the students demonstrated better logical analysis, outdoing both chatbots. In updating command condition, ChatGPT Plus displayed good causal reasoning ability while ChatGPT kept unchanged. Our study reveals that human readers and ChatGPTs have their respective advantages and disadvantages in drawing inferences from text reading comprehension, unlocking a complementary relationship in text-based reasoning.

We propose controlled decoding (CD), a novel off-policy reinforcement learning method to control the autoregressive generation from language models towards high reward outcomes. CD solves an off-policy reinforcement learning problem through a value function for the reward, which we call a prefix scorer. The prefix scorer is used at inference time to steer the generation towards higher reward outcomes. We show that the prefix scorer may be trained on (possibly) off-policy data to predict the expected reward when decoding is continued from a partially decoded response. We empirically demonstrate that CD is effective as a control mechanism on Reddit conversations corpus. We also show that the modularity of the design of CD makes it possible to control for multiple rewards, effectively solving a multi-objective reinforcement learning problem with no additional complexity. Finally, we show that CD can be applied in a novel blockwise fashion at inference-time, again without the need for any training-time changes, essentially bridging the gap between the popular best-of-$K$ strategy and token-level reinforcement learning. This makes CD a promising approach for alignment of language models.

Policy gradient lies at the core of deep reinforcement learning (RL) in continuous domains. Despite much success, it is often observed in practice that RL training with policy gradient can fail for many reasons, even on standard control problems with known solutions. We propose a framework for understanding one inherent limitation of the policy gradient approach: the optimization landscape in the policy space can be extremely non-smooth or fractal for certain classes of MDPs, such that there does not exist gradient to be estimated in the first place. We draw on techniques from chaos theory and non-smooth analysis, and analyze the maximal Lyapunov exponents and H\"older exponents of the policy optimization objectives. Moreover, we develop a practical method that can estimate the local smoothness of objective function from samples to identify when the training process has encountered fractal landscapes. We show experiments to illustrate how some failure cases of policy optimization can be explained by such fractal landscapes.

Large language models (LLMs), with their remarkable conversational capabilities, have demonstrated impressive performance across various applications and have emerged as formidable AI assistants. In view of this, it raises an intuitive question: Can we harness the power of LLMs to build multimodal AI assistants for visual applications? Recently, several multi-modal models have been developed for this purpose. They typically pre-train an adaptation module to align the semantics of the vision encoder and language model, followed by fine-tuning on instruction-following data. However, despite the success of this pipeline in image and language understanding, its effectiveness in joint video and language understanding has not been widely explored. In this paper, we aim to develop a novel multi-modal foundation model capable of comprehending video, image, and language within a general framework. To achieve this goal, we introduce Valley, a Video Assistant with Large Language model Enhanced abilitY. The Valley consists of a LLM, a temporal modeling module, a visual encoder, and a simple projection module designed to bridge visual and textual modes. To empower Valley with video comprehension and instruction-following capabilities, we construct a video instruction dataset and adopt a two-stage tuning procedure to train it. Specifically, we employ ChatGPT to facilitate the construction of task-oriented conversation data encompassing various tasks, including multi-shot captions, long video descriptions, action recognition, causal relationship inference, etc. Subsequently, we adopt a pre-training-then-instructions-tuned pipeline to align visual and textual modalities and improve the instruction-following capability of Valley. Qualitative experiments demonstrate that Valley has the potential to function as a highly effective video assistant that can make complex video understanding scenarios easy.