Multimodal learning, in the context of machine learning, is a type of deep learning using multiple modalities of data, such as text, audio, or images.
In contrast, unimodal models can process only one type of data, such as text (typically represented as feature vectors) or images. Multimodal learning is different from combining unimodal models trained independently. It combines information from different modalities in order to make better predictions.[1]
Large multimodal models, such as Google Gemini and GPT-4o, have become increasingly popular since 2023, enabling increased versatility and a broader understanding of real-world phenomena.[2]
Motivation
Data usually comes with different modalities which carry different information. For example, it is very common to caption an image to convey the information not presented in the image itself. Similarly, sometimes it is more straightforward to use an image to describe information which may not be obvious from text. As a result, if different words appear in similar images, then these words likely describe the same thing. Conversely, if a word is used to describe seemingly dissimilar images, then these images may represent the same object. Thus, in cases dealing with multi-modal data, it is important to use a model which is able to jointly represent the information such that the model can capture the combined information from different modalities.
Multimodal transformers
Transformers can also be used/adapted for modalities (input or output) beyond just text, usually by finding a way to "tokenize" the modality.
Multimodal models can either be trained from scratch, or by finetuning. A 2022 study found that Transformers pretrained only on natural language can be finetuned on only 0.03% of parameters and become competitive with LSTMs on a variety of logical and visual tasks, demonstrating transfer learning.[3] The LLaVA was a vision-language model composed of a language model (Vicuna-13B)[4] and a vision model (ViT-L/14), connected by a linear layer. Only the linear layer is finetuned.[5]
Vision transformers[6] adapt the transformer to computer vision by breaking down input images as a series of patches, turning them into vectors, and treating them like tokens in a standard transformer.
Conformer[7] and later Whisper[8] follow the same pattern for speech recognition, first turning the speech signal into a spectrogram, which is then treated like an image, i.e. broken down into a series of patches, turned into vectors and treated like tokens in a standard transformer.
Perceivers[9][10] are a variant of Transformers designed for multimodality.
For image generation, notable architectures are DALL-E 1 (2021), Parti (2022),[11] and Muse (2023).[12] Unlike later models, DALL-E is not a diffusion model. Instead, it uses a decoder-only Transformer that autoregressively generates a text, followed by the token representation of an image, which is then converted by a variational autoencoder to an image.[13] Parti is an encoder-decoder Transformer, where the encoder processes a text prompt, and the decoder generates a token representation of an image.[14] Muse is an encoder-only Transformer that is trained to predict masked image tokens from unmasked image tokens. During generation, all input tokens are masked, and the highest-confidence predictions are included for the next iteration, until all tokens are predicted.[12]Multimodal large language models
Multimodality means "having several modalities", and a "modality" refers to a type of input or output, such as video, image, audio, text, proprioception, etc.[15] There have been many AI models trained specifically to ingest one modality and output another modality, such as AlexNet for image to label,[16] visual question answering for image-text to text,[17] and speech recognition for speech to text.
A common method to create multimodal models out of an LLM is to "tokenize" the output of a trained encoder. Concretely, one can construct an LLM that can understand images as follows: take a trained LLM, and take a trained image encoder . Make a small multilayered perceptron , so that for any image , the post-processed vector has the same dimensions as an encoded token. That is an "image token". Then, one can interleave text tokens and image tokens. The compound model is then fine-tuned on an image-text dataset. This basic construction can be applied with more sophistication to improve the model. The image encoder may be frozen to improve stability.[18]
Flamingo demonstrated the effectiveness of the tokenization method, finetuning a pair of pretrained language model and image encoder to perform better on visual question answering than models trained from scratch.[19] Google PaLM model was fine-tuned into a multimodal model PaLM-E using the tokenization method, and applied to robotic control.[20] LLaMA models have also been turned multimodal using the tokenization method, to allow image inputs,[21] and video inputs.[22]
GPT-4 can use both text and image as inputs[23] (although the vision component was not released to the public until GPT-4V[24]); Google DeepMind's Gemini is also multimodal.[25] Mistral introduced its own multimodel Pixtral 12B model in September 2024.[26]Multimodal deep Boltzmann machines
A Boltzmann machine is a type of stochastic neural network invented by Geoffrey Hinton and Terry Sejnowski in 1985. Boltzmann machines can be seen as the stochastic, generative counterpart of Hopfield nets. They are named after the Boltzmann distribution in statistical mechanics. The units in Boltzmann machines are divided into two groups: visible units and hidden units. Each unit is like a neuron with a binary output that represents whether it's activated or not.[27] General Boltzmann machines allow connection between any units. However, learning is impractical using general Boltzmann Machines because the computational time is exponential to the size of the machine[citation needed]. A more efficient architecture is called restricted Boltzmann machine where connection is only allowed between hidden unit and visible unit, which is described in the next section.
Multimodal deep Boltzmann machines can process and learn from different types of information, such as images and text, simultaneously. This can notably be done by having a separate deep Boltzmann machine for each modality, for example one for images and one for text, joined at an additional top hidden layer.[28]
Application
Multimodal deep Boltzmann machines are successfully used in classification and missing data retrieval. The classification accuracy of multimodal deep Boltzmann machine outperforms support vector machines, latent Dirichlet allocation and deep belief network, when models are tested on data with both image-text modalities or with single modality.[citation needed] Multimodal deep Boltzmann machines are also able to predict missing modalities given the observed ones with reasonably good precision.[citation needed] Self-supervised learning brings a more interesting and powerful model for multimodality. OpenAI developed CLIP and DALL-E models that revolutionized multimodality.
Multimodal deep learning is used for cancer screening – at least one system under development integrates such different types of data.[29][30]
See also
References
Wikiwand in your browser!
Seamless Wikipedia browsing. On steroids.
Every time you click a link to Wikipedia, Wiktionary or Wikiquote in your browser's search results, it will show the modern Wikiwand interface.
Wikiwand extension is a five stars, simple, with minimum permission required to keep your browsing private, safe and transparent.
