Natural language processing

Subterms

    Latest story

    63 Shares159 Views

    Leveraging language to understand machines

    by

    Natural language conveys ideas, actions, information, and intent through context and syntax; further, there are volumes of it contained in databases. This makes it an excellent source of data to train machine-learning systems on. Two master’s of engineering students in the 6A MEng Thesis Program at MIT, Irene Terpstra ’23 and Rujul Gandhi ’22, are working with mentors in the MIT-IBM Watson AI Lab to use this power of natural language to build AI systems.

    As computing is becoming more advanced, researchers are looking to improve the hardware that they run on; this means innovating to create new computer chips. And, since there is literature already available on modifications that can be made to achieve certain parameters and performance, Terpstra and her mentors and advisors Anantha Chandrakasan, MIT School of Engineering dean and the Vannevar Bush Professor of Electrical Engineering and Computer Science, and IBM’s researcher Xin Zhang, are developing an AI algorithm that assists in chip design.

    “I’m creating a workflow to systematically analyze how these language models can help the circuit design process. What reasoning powers do they have, and how can it be integrated into the chip design process?” says Terpstra. “And then on the other side, if that proves to be useful enough, [we’ll] see if they can automatically design the chips themselves, attaching it to a reinforcement learning algorithm.”

    To do this, Terpstra’s team is creating an AI system that can iterate on different designs. It means experimenting with various pre-trained large language models (like ChatGPT, Llama 2, and Bard), using an open-source circuit simulator language called NGspice, which has the parameters of the chip in code form, and a reinforcement learning algorithm. With text prompts, researchers will be able to query how the physical chip should be modified to achieve a certain goal in the language model and produced guidance for adjustments. This is then transferred into a reinforcement learning algorithm that updates the circuit design and outputs new physical parameters of the chip.

    “The final goal would be to combine the reasoning powers and the knowledge base that is baked into these large language models and combine that with the optimization power of the reinforcement learning algorithms and have that design the chip itself,” says Terpstra.

    Rujul Gandhi works with the raw language itself. As an undergraduate at MIT, Gandhi explored linguistics and computer sciences, putting them together in her MEng work. “I’ve been interested in communication, both between just humans and between humans and computers,” Gandhi says.

    Robots or other interactive AI systems are one area where communication needs to be understood by both humans and machines. Researchers often write instructions for robots using formal logic. This helps ensure that commands are being followed safely and as intended, but formal logic can be difficult for users to understand, while natural language comes easily. To ensure this smooth communication, Gandhi and her advisors Yang Zhang of IBM and MIT assistant professor Chuchu Fan are building a parser that converts natural language instructions into a machine-friendly form. Leveraging the linguistic structure encoded by the pre-trained encoder-decoder model T5, and a dataset of annotated, basic English commands for performing certain tasks, Gandhi’s system identifies the smallest logical units, or atomic propositions, which are present in a given instruction.

    “Once you’ve given your instruction, the model identifies all the smaller sub-tasks you want it to carry out,” Gandhi says. “Then, using a large language model, each sub-task can be compared against the available actions and objects in the robot’s world, and if any sub-task can’t be carried out because a certain object is not recognized, or an action is not possible, the system can stop right there to ask the user for help.”

    This approach of breaking instructions into sub-tasks also allows her system to understand logical dependencies expressed in English, like, “do task X until event Y happens.” Gandhi uses a dataset of step-by-step instructions across robot task domains like navigation and manipulation, with a focus on household tasks. Using data that are written just the way humans would talk to each other has many advantages, she says, because it means a user can be more flexible about how they phrase their instructions.

    Another of Gandhi’s projects involves developing speech models. In the context of speech recognition, some languages are considered “low resource” since they might not have a lot of transcribed speech available, or might not have a written form at all. “One of the reasons I applied to this internship at the MIT-IBM Watson AI Lab was an interest in language processing for low-resource languages,” she says. “A lot of language models today are very data-driven, and when it’s not that easy to acquire all of that data, that’s when you need to use the limited data efficiently.” 

    Speech is just a stream of sound waves, but humans having a conversation can easily figure out where words and thoughts start and end. In speech processing, both humans and language models use their existing vocabulary to recognize word boundaries and understand the meaning. In low- or no-resource languages, a written vocabulary might not exist at all, so researchers can’t provide one to the model. Instead, the model can make note of what sound sequences occur together more frequently than others, and infer that those might be individual words or concepts. In Gandhi’s research group, these inferred words are then collected into a pseudo-vocabulary that serves as a labeling method for the low-resource language, creating labeled data for further applications.

    The applications for language technology are “pretty much everywhere,” Gandhi says. “You could imagine people being able to interact with software and devices in their native language, their native dialect. You could imagine improving all the voice assistants that we use. You could imagine it being used for translation or interpretation.” More

    More stories

    • 88 Shares99 Views

      in Data Management & Statistics

      MIT researchers make language models scalable self-learners

      by

      Socrates once said: “It is not the size of a thing, but the quality that truly matters. For it is in the nature of substance, not its volume, that true value is found.”

      Does size always matter for large language models (LLMs)? In a technological landscape bedazzled by LLMs taking center stage, a team of MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) researchers think smaller models shouldn’t be overlooked, especially for natural language understanding products widely deployed in the industry.

      To that end, the researchers cooked up an approach to long-standing problems of inefficiency and privacy associated with big, text-based AI models — a logic-aware model that outperforms 500-times-bigger counterparts on some language understanding tasks without human-generated annotations, while preserving privacy and robustness with high performance.

      LLMs, which have shown some promising skills in generating language, art, and code, are computationally expensive, and their data requirements can risk privacy leaks when using application programming interfaces for data upload. Smaller models have been historically less capable, particularly in multitasking and weakly supervised tasks, compared to their larger counterparts.

      So what’s helping these smaller models act so mighty, then? Something called “textual entailment,” a way to help these models understand a variety of language tasks, where if one sentence (the premise) is true, then the other sentence (the hypothesis) is likely to be true as well. For example, if the premise is, “all cats have tails” then the hypothesis “a tabby cat has a tail” would be entailed by the premise. This concept is used to train an “entailment model” that proved to be less biased than other language models, from the team’s previous research. They then created “prompts” that the models can use to figure out if certain information is entailed by a given sentence or phrase according to different tasks. This method improved the model’s ability to adapt to different tasks without any additional training, known as zero-shot adaptation.

      In the realm of “natural language understanding,” there are various applications that hinge on determining the relationship between two pieces of text. For example, in sentiment classification, a statement like “I think the movie is good” can be inferred or entailed from a movie review that says, “I like the story and the acting is great,” indicating a positive sentiment. Another is news classification, where the topic of a news article can be inferred from its content. For example, a statement like “the news article is about sports” can be entailed if the main content of the article reports on an NBA game. The key insight was that many existing natural language understanding tasks could be recast as an entailment (i.e., logical inference in natural language) task. 

      “Our research is about improving the ability of computer programs to understand and process natural language — the way humans speak and write. Our self-trained, 350-million-parameter entailment models, without human-generated labels, outperform supervised language models with 137 to 175 billion parameters,” says MIT CSAIL postdoc Hongyin Luo, lead author on a new paper about the study. “This has potential to reshape the landscape of AI and machine learning, providing a more scalable, trustworthy, and cost-effective solution to language modeling,” says Luo. “By proving that smaller models can perform at the same level as larger ones for language understanding, this work paves the way for more sustainable and privacy-preserving AI technologies.” 

      The team discovered that they could improve the model’s performance even more by using a technique called “self-training,” where the model uses its own predictions to teach itself, effectively learning without human supervision and additional annotated training data.The self-training method significantly improved performance on a bunch of downstream tasks, including sentiment analysis, question-answering, and news classification. It outperformed both Google’s LaMDA and FLAN in zero-shot capabilities, GPT models, and other supervised algorithms. 

      However, one challenge with self-training is that the model can sometimes generate incorrect or noisy labels that harm performance. To overcome this, they developed a new algorithm called ‘SimPLE’ (Simple Pseudo-Label Editing), a process to review and modify the pseudo-labels made in initial rounds of learning. By correcting any mislabeled instances, it improved the overall quality of the self-generated labels. This not only made the models more effective at understanding language, but more robust when faced with adversarial data. 

      As with most research, there are some limitations. The self-training on multi-class classification tasks didn’t perform as well as on binary natural language understanding tasks, indicating the challenge of applying entailment models to multi-choice tasks.“This research presents an efficient and effective way to train large language models (LLMs) by formulating natural language understanding tasks as contextual entailment problems and employing a pseudo-labeling self-training mechanism to incorporate large quantities of unlabelled text data in the training process,” adds CSAIL Senior Research Scientist James Glass, who is also an author on the paper. “While the field of LLMs is undergoing rapid and dramatic changes, this research shows that it is possible to produce relatively compact language models that perform very well on benchmark understanding tasks compared to their peers of roughly the same size, or even much larger language models.”

      “Entailment task is a popular proxy to evaluate “understanding” of a given context by an AI model,” says Leonid Karlinsky, research staff member at the MIT-IBM Watson AI Lab. “It is used in many areas analyzing models with unimodal, like LLMs, and and multi-modal, like VLMs [visual language models] inputs, simplifying the task of question-answering about a given input context to a binary classification problem — does this context entail a certain (e.g., text) conclusion or not? This paper makes two contributions in this space. First, it proposes a way to improve the zero-shot (without additional tuning) NLU performance and robustness to adversarial attacks via tuning with synthesized (specialized) entailment tasks generated for the primal NLU task. Second, it offers a self-supervised SimPLE method including pseudo-labeling and confidence-based filtering to further improve large LLMs’ NLU performance.”

      Luo and Glass wrote the paper with Yoon Kim, a CSAIL member and assistant professor in MIT’s Department of Electrical Engineering and Computer Science, and Jiaxin Ge of Peking University. Their work will be presented at the meeting of the Association for Computational Linguistics in Toronto, Ontario this July. This research was supported by a grant from the Hong Kong Innovation AI program. More

    • 125 Shares99 Views

      in Data Management & Statistics

      Large language models are biased. Can logic help save them?

      by

      Turns out, even language models “think” they’re biased. When prompted in ChatGPT, the response was as follows: “Yes, language models can have biases, because the training data reflects the biases present in society from which that data was collected. For example, gender and racial biases are prevalent in many real-world datasets, and if a language model is trained on that, it can perpetuate and amplify these biases in its predictions.” A well-known but dangerous problem. 

      Humans (typically) can dabble with both logical and stereotypical reasoning when learning. Still, language models mainly mimic the latter, an unfortunate narrative we’ve seen play out ad nauseam when the ability to employ reasoning and critical thinking is absent. So would injecting logic into the fray be enough to mitigate such behavior? 

      Scientists from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) had an inkling that it might, so they set off to examine if logic-aware language models could significantly avoid more harmful stereotypes. They trained a language model to predict the relationship between two sentences, based on context and semantic meaning, using a dataset with labels for text snippets detailing if a second phrase “entails,” “contradicts,” or is neutral with respect to the first one. Using this dataset — natural language inference — they found that the newly trained models were significantly less biased than other baselines, without any extra data, data editing, or additional training algorithms.

      For example, with the premise “the person is a doctor” and the hypothesis “the person is masculine,” using these logic-trained models, the relationship would be classified as “neutral,” since there’s no logic that says the person is a man. With more common language models, two sentences might seem to be correlated due to some bias in training data, like “doctor” might be pinged with “masculine,” even when there’s no evidence that the statement is true. 

      At this point, the omnipresent nature of language models is well-known: Applications in natural language processing, speech recognition, conversational AI, and generative tasks abound. While not a nascent field of research, growing pains can take a front seat as they increase in complexity and capability. 

      “Current language models suffer from issues with fairness, computational resources, and privacy,” says MIT CSAIL postdoc Hongyin Luo, the lead author of a new paper about the work. “Many estimates say that the CO2 emission of training a language model can be higher than the lifelong emission of a car. Running these large language models is also very expensive because of the amount of parameters and the computational resources they need. With privacy, state-of-the-art language models developed by places like ChatGPT or GPT-3 have their APIs where you must upload your language, but there’s no place for sensitive information regarding things like health care or finance. To solve these challenges, we proposed a logical language model that we qualitatively measured as fair, is 500 times smaller than the state-of-the-art models, can be deployed locally, and with no human-annotated training samples for downstream tasks. Our model uses 1/400 the parameters compared with the largest language models, has better performance on some tasks, and significantly saves computation resources.” 

      This model, which has 350 million parameters, outperformed some very large-scale language models with 100 billion parameters on logic-language understanding tasks. The team evaluated, for example, popular BERT pretrained language models with their “textual entailment” ones on stereotype, profession, and emotion bias tests. The latter outperformed other models with significantly lower bias, while preserving the language modeling ability. The “fairness” was evaluated with something called ideal context association (iCAT) tests, where higher iCAT scores mean fewer stereotypes. The model had higher than 90 percent iCAT scores, while other strong language understanding models ranged between 40 to 80. 

      Luo wrote the paper alongside MIT Senior Research Scientist James Glass. They will present the work at the Conference of the European Chapter of the Association for Computational Linguistics in Croatia. 

      Unsurprisingly, the original pretrained language models the team examined were teeming with bias, confirmed by a slew of reasoning tests demonstrating how professional and emotion terms are significantly biased to the feminine or masculine words in the gender vocabulary. 

      With professions, a language model (which is biased) thinks that “flight attendant,” “secretary,” and “physician’s assistant” are feminine jobs, while “fisherman,” “lawyer,” and “judge” are masculine. Concerning emotions, a language model thinks that “anxious,” “depressed,” and “devastated” are feminine.

      While we may still be far away from a neutral language model utopia, this research is ongoing in that pursuit. Currently, the model is just for language understanding, so it’s based on reasoning among existing sentences. Unfortunately, it can’t generate sentences for now, so the next step for the researchers would be targeting the uber-popular generative models built with logical learning to ensure more fairness with computational efficiency. 

      “Although stereotypical reasoning is a natural part of human recognition, fairness-aware people conduct reasoning with logic rather than stereotypes when necessary,” says Luo. “We show that language models have similar properties. A language model without explicit logic learning makes plenty of biased reasoning, but adding logic learning can significantly mitigate such behavior. Furthermore, with demonstrated robust zero-shot adaptation ability, the model can be directly deployed to different tasks with more fairness, privacy, and better speed.” More

    • 88 Shares129 Views

      in Data Management & Statistics

      Large language models help decipher clinical notes

      by

      Electronic health records (EHRs) need a new public relations manager. Ten years ago, the U.S. government passed a law that required hospitals to digitize their health records with the intent of improving and streamlining care. The enormous amount of information in these now-digital records could be used to answer very specific questions beyond the scope of clinical trials: What’s the right dose of this medication for patients with this height and weight? What about patients with a specific genomic profile?

      Unfortunately, most of the data that could answer these questions is trapped in doctor’s notes, full of jargon and abbreviations. These notes are hard for computers to understand using current techniques — extracting information requires training multiple machine learning models. Models trained for one hospital, also, don’t work well at others, and training each model requires domain experts to label lots of data, a time-consuming and expensive process. 

      An ideal system would use a single model that can extract many types of information, work well at multiple hospitals, and learn from a small amount of labeled data. But how? Researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) believed that to disentangle the data, they needed to call on something bigger: large language models. To pull that important medical information, they used a very big, GPT-3 style model to do tasks like expand overloaded jargon and acronyms and extract medication regimens. 

      For example, the system takes an input, which in this case is a clinical note, “prompts” the model with a question about the note, such as “expand this abbreviation, C-T-A.” The system returns an output such as “clear to auscultation,” as opposed to say, a CT angiography. The objective of extracting this clean data, the team says, is to eventually enable more personalized clinical recommendations. 

      Medical data is, understandably, a pretty tricky resource to navigate freely. There’s plenty of red tape around using public resources for testing the performance of large models because of data use restrictions, so the team decided to scrape together their own. Using a set of short, publicly available clinical snippets, they cobbled together a small dataset to enable evaluation of the extraction performance of large language models. 

      “It’s challenging to develop a single general-purpose clinical natural language processing system that will solve everyone’s needs and be robust to the huge variation seen across health datasets. As a result, until today, most clinical notes are not used in downstream analyses or for live decision support in electronic health records. These large language model approaches could potentially transform clinical natural language processing,” says David Sontag, MIT professor of electrical engineering and computer science, principal investigator in CSAIL and the Institute for Medical Engineering and Science, and supervising author on a paper about the work, which will be presented at the Conference on Empirical Methods in Natural Language Processing. “The research team’s advances in zero-shot clinical information extraction makes scaling possible. Even if you have hundreds of different use cases, no problem — you can build each model with a few minutes of work, versus having to label a ton of data for that particular task.”

      For example, without any labels at all, the researchers found these models could achieve 86 percent accuracy at expanding overloaded acronyms, and the team developed additional methods to boost this further to 90 percent accuracy, with still no labels required.

      Imprisoned in an EHR 

      Experts have been steadily building up large language models (LLMs) for quite some time, but they burst onto the mainstream with GPT-3’s widely covered ability to complete sentences. These LLMs are trained on a huge amount of text from the internet to finish sentences and predict the next most likely word. 

      While previous, smaller models like earlier GPT iterations or BERT have pulled off a good performance for extracting medical data, they still require substantial manual data-labeling effort. 

      For example, a note, “pt will dc vanco due to n/v” means that this patient (pt) was taking the antibiotic vancomycin (vanco) but experienced nausea and vomiting (n/v) severe enough for the care team to discontinue (dc) the medication. The team’s research avoids the status quo of training separate machine learning models for each task (extracting medication, side effects from the record, disambiguating common abbreviations, etc). In addition to expanding abbreviations, they investigated four other tasks, including if the models could parse clinical trials and extract detail-rich medication regimens.  

      “Prior work has shown that these models are sensitive to the prompt’s precise phrasing. Part of our technical contribution is a way to format the prompt so that the model gives you outputs in the correct format,” says Hunter Lang, CSAIL PhD student and author on the paper. “For these extraction problems, there are structured output spaces. The output space is not just a string. It can be a list. It can be a quote from the original input. So there’s more structure than just free text. Part of our research contribution is encouraging the model to give you an output with the correct structure. That significantly cuts down on post-processing time.”

      The approach can’t be applied to out-of-the-box health data at a hospital: that requires sending private patient information across the open internet to an LLM provider like OpenAI. The authors showed that it’s possible to work around this by distilling the model into a smaller one that could be used on-site.

      The model — sometimes just like humans — is not always beholden to the truth. Here’s what a potential problem might look like: Let’s say you’re asking the reason why someone took medication. Without proper guardrails and checks, the model might just output the most common reason for that medication, if nothing is explicitly mentioned in the note. This led to the team’s efforts to force the model to extract more quotes from data and less free text.

      Future work for the team includes extending to languages other than English, creating additional methods for quantifying uncertainty in the model, and pulling off similar results with open-sourced models. 

      “Clinical information buried in unstructured clinical notes has unique challenges compared to general domain text mostly due to large use of acronyms, and inconsistent textual patterns used across different health care facilities,” says Sadid Hasan, AI lead at Microsoft and former executive director of AI at CVS Health, who was not involved in the research. “To this end, this work sets forth an interesting paradigm of leveraging the power of general domain large language models for several important zero-/few-shot clinical NLP tasks. Specifically, the proposed guided prompt design of LLMs to generate more structured outputs could lead to further developing smaller deployable models by iteratively utilizing the model generated pseudo-labels.”

      “AI has accelerated in the last five years to the point at which these large models can predict contextualized recommendations with benefits rippling out across a variety of domains such as suggesting novel drug formulations, understanding unstructured text, code recommendations or create works of art inspired by any number of human artists or styles,” says Parminder Bhatia, who was formerly Head of Machine Learning at AWS Health AI and is currently Head of ML for low-code applications leveraging large language models at AWS AI Labs. “One of the applications of these large models [the team has] recently launched is Amazon CodeWhisperer, which is [an] ML-powered coding companion that helps developers in building applications.”

      As part of the MIT Abdul Latif Jameel Clinic for Machine Learning in Health, Agrawal, Sontag, and Lang wrote the paper alongside Yoon Kim, MIT assistant professor and CSAIL principal investigator, and Stefan Hegselmann, a visiting PhD student from the University of Muenster. First-author Agrawal’s research was supported by a Takeda Fellowship, the MIT Deshpande Center for Technological Innovation, and the MLA@CSAIL Initiatives. More

    • 100 Shares189 Views

      in Data Management & Statistics

      New leadership at MIT’s Center for Biomedical Innovation

      by

      As it continues in its mission to improve global health through the development and implementation of biomedical innovation, the MIT Center for Biomedical Innovation (CBI) today announced changes to its leadership team: Stacy Springs has been named executive director, and Professor Richard Braatz has joined as the center’s new associate faculty director.

      The change in leadership comes at a time of rapid development in new therapeutic modalities, growing concern over global access to biologic medicines and healthy food, and widespread interest in applying computational tools and multi-disciplinary approaches to address long-standing biomedical challenges.

      “This marks an exciting new chapter for the CBI,” says faculty director Anthony J. Sinskey, professor of biology, who cofounded CBI in 2005. “As I look back at almost 20 years of CBI history, I see an exponential growth in our activities, educational offerings, and impact.”

      The center’s collaborative research model accelerates innovation in biotechnology and biomedical research, drawing on the expertise of faculty and researchers in MIT’s schools of Engineering and Science, the MIT Schwarzman College of Computing, and the MIT Sloan School of Management.

      Springs steps into the role of executive director having previously served as senior director of programs for CBI and as executive director of CBI’s Biomanufacturing Program and its Consortium on Adventitious Agent Contamination in Biomanufacturing (CAACB). She succeeds Gigi Hirsch, who founded the NEW Drug Development ParadIGmS (NEWDIGS) Initiative at CBI in 2009. Hirsch and NEWDIGS have now moved to Tufts Medical Center, establishing a headquarters at the new Center for Biomedical System Design within the Institute for Clinical Research and Health Policy Studies there.

      Braatz, a chemical engineer whose work is informed by mathematical modeling and computational techniques, conducts research in process data analytics, design, and control of advanced manufacturing systems.

      “It’s been great to interact with faculty from across the Institute who have complementary expertise,” says Braatz, the Edwin R. Gilliland Professor in the Department of Chemical Engineering. “Participating in CBI’s workshops has led to fruitful partnerships with companies in tackling industry-wide challenges.”

      CBI is housed under the Institute for Data Systems and Society and, specifically, the Sociotechnical Systems Research Center in the MIT Schwarzman College of Computing. CBI is home to two biomanufacturing consortia: the CAACB and the Biomanufacturing Consortium (BioMAN). Through these precompetitive collaborations, CBI researchers work with biomanufacturers and regulators to advance shared interests in biomanufacturing.

      In addition, CBI researchers are engaged in several sponsored research programs focused on integrated continuous biomanufacturing capabilities for monoclonal antibodies and vaccines, analytical technologies to measure quality and safety attributes of a variety of biologics, including gene and cell therapies, and rapid-cycle development of virus-like particle vaccines for SARS-CoV-2.

      In another significant initiative, CBI researchers are applying data analytics strategies to biomanufacturing problems. “In our smart data analytics project, we are creating new decision support tools and algorithms for biomanufacturing process control and plant-level decision-making. Further, we are leveraging machine learning and natural language processing to improve post-market surveillance studies,” says Springs.

      CBI is also working on advanced manufacturing for cell and gene therapies, among other new modalities, and is a part of the Singapore-MIT Alliance for Research and Technology – Critical Analytics for Manufacturing Personalized-Medicine (SMART CAMP). SMART CAMP is an international research effort focused on developing the analytical tools and biological understanding of critical quality attributes that will enable the manufacture and delivery of improved cell therapies to patients.

      “This is a crucial time for biomanufacturing and for innovation across the health-care value chain. The collaborative efforts of MIT researchers and consortia members will drive fundamental discovery and inform much-needed progress in industry,” says MIT Vice President for Research Maria Zuber.

      “CBI has a track record of engaging with health-care ecosystem challenges. I am confident that under the new leadership, it will continue to inspire MIT, the United States, and the entire world to improve the health of all people,” adds Daniel Huttenlocher, dean of the MIT Schwarzman College of Computing. More

    • 63 Shares189 Views

      in Data Management & Statistics

      Hallucinating to better text translation

      by

      As babies, we babble and imitate our way to learning languages. We don’t start off reading raw text, which requires fundamental knowledge and understanding about the world, as well as the advanced ability to interpret and infer descriptions and relationships. Rather, humans begin our language journey slowly, by pointing and interacting with our environment, basing our words and perceiving their meaning through the context of the physical and social world. Eventually, we can craft full sentences to communicate complex ideas.

      Similarly, when humans begin learning and translating into another language, the incorporation of other sensory information, like multimedia, paired with the new and unfamiliar words, like flashcards with images, improves language acquisition and retention. Then, with enough practice, humans can accurately translate new, unseen sentences in context without the accompanying media; however, imagining a picture based on the original text helps.

      This is the basis of a new machine learning model, called VALHALLA, by researchers from MIT, IBM, and the University of California at San Diego, in which a trained neural network sees a source sentence in one language, hallucinates an image of what it looks like, and then uses both to translate into a target language. The team found that their method demonstrates improved accuracy of machine translation over text-only translation. Further, it provided an additional boost for cases with long sentences, under-resourced languages, and instances where part of the source sentence is inaccessible to the machine translator.

      As a core task within the AI field of natural language processing (NLP), machine translation is an “eminently practical technology that’s being used by millions of people every day,” says study co-author Yoon Kim, assistant professor in MIT’s Department of Electrical Engineering and Computer Science with affiliations in the Computer Science and Artificial Intelligence Laboratory (CSAIL) and the MIT-IBM Watson AI Lab. With recent, significant advances in deep learning, “there’s been an interesting development in how one might use non-text information — for example, images, audio, or other grounding information — to tackle practical tasks involving language” says Kim, because “when humans are performing language processing tasks, we’re doing so within a grounded, situated world.” The pairing of hallucinated images and text during inference, the team postulated, imitates that process, providing context for improved performance over current state-of-the-art techniques, which utilize text-only data.

      This research will be presented at the IEEE / CVF Computer Vision and Pattern Recognition Conference this month. Kim’s co-authors are UC San Diego graduate student Yi Li and Professor Nuno Vasconcelos, along with research staff members Rameswar Panda, Chun-fu “Richard” Chen, Rogerio Feris, and IBM Director David Cox of IBM Research and the MIT-IBM Watson AI Lab.

      Learning to hallucinate from images

      When we learn new languages and to translate, we’re often provided with examples and practice before venturing out on our own. The same is true for machine-translation systems; however, if images are used during training, these AI methods also require visual aids for testing, limiting their applicability, says Panda.

      “In real-world scenarios, you might not have an image with respect to the source sentence. So, our motivation was basically: Instead of using an external image during inference as input, can we use visual hallucination — the ability to imagine visual scenes — to improve machine translation systems?” says Panda.

      To do this, the team used an encoder-decoder architecture with two transformers, a type of neural network model that’s suited for sequence-dependent data, like language, that can pay attention key words and semantics of a sentence. One transformer generates a visual hallucination, and the other performs multimodal translation using outputs from the first transformer.

      During training, there are two streams of translation: a source sentence and a ground-truth image that is paired with it, and the same source sentence that is visually hallucinated to make a text-image pair. First the ground-truth image and sentence are tokenized into representations that can be handled by transformers; for the case of the sentence, each word is a token. The source sentence is tokenized again, but this time passed through the visual hallucination transformer, outputting a hallucination, a discrete image representation of the sentence. The researchers incorporated an autoregression that compares the ground-truth and hallucinated representations for congruency — e.g., homonyms: a reference to an animal “bat” isn’t hallucinated as a baseball bat. The hallucination transformer then uses the difference between them to optimize its predictions and visual output, making sure the context is consistent.

      The two sets of tokens are then simultaneously passed through the multimodal translation transformer, each containing the sentence representation and either the hallucinated or ground-truth image. The tokenized text translation outputs are compared with the goal of being similar to each other and to the target sentence in another language. Any differences are then relayed back to the translation transformer for further optimization.

      For testing, the ground-truth image stream drops off, since images likely wouldn’t be available in everyday scenarios.

      “To the best of our knowledge, we haven’t seen any work which actually uses a hallucination transformer jointly with a multimodal translation system to improve machine translation performance,” says Panda.

      Visualizing the target text

      To test their method, the team put VALHALLA up against other state-of-the-art multimodal and text-only translation methods. They used public benchmark datasets containing ground-truth images with source sentences, and a dataset for translating text-only news articles. The researchers measured its performance over 13 tasks, ranging from translation on well-resourced languages (like English, German, and French), under-resourced languages (like English to Romanian) and non-English (like Spanish to French). The group also tested varying transformer model sizes, how accuracy changes with the sentence length, and translation under limited textual context, where portions of the text were hidden from the machine translators.

      The team observed significant improvements over text-only translation methods, improving data efficiency, and that smaller models performed better than the larger base model. As sentences became longer, VALHALLA’s performance over other methods grew, which the researchers attributed to the addition of more ambiguous words. In cases where part of the sentence was masked, VALHALLA could recover and translate the original text, which the team found surprising.

      Further unexpected findings arose: “Where there weren’t as many training [image and] text pairs, [like for under-resourced languages], improvements were more significant, which indicates that grounding in images helps in low-data regimes,” says Kim. “Another thing that was quite surprising to me was this improved performance, even on types of text that aren’t necessarily easily connectable to images. For example, maybe it’s not so surprising if this helps in translating visually salient sentences, like the ‘there is a red car in front of the house.’ [However], even in text-only [news article] domains, the approach was able to improve upon text-only systems.”

      While VALHALLA performs well, the researchers note that it does have limitations, requiring pairs of sentences to be annotated with an image, which could make it more expensive to obtain. It also performs better in its ground domain and not the text-only news articles. Moreover, Kim and Panda note, a technique like VALHALLA is still a black box, with the assumption that hallucinated images are providing helpful information, and the team plans to investigate what and how the model is learning in order to validate their methods.

      In the future, the team plans to explore other means of improving translation. “Here, we only focus on images, but there are other types of a multimodal information — for example, speech, video or touch, or other sensory modalities,” says Panda. “We believe such multimodal grounding can lead to even more efficient machine translation models, potentially benefiting translation across many low-resource languages spoken in the world.”

      This research was supported, in part, by the MIT-IBM Watson AI Lab and the National Science Foundation. More

    • 113 Shares169 Views

      in Data Management & Statistics

      Artificial intelligence predicts patients’ race from their medical images

      by

      The miseducation of algorithms is a critical problem; when artificial intelligence mirrors unconscious thoughts, racism, and biases of the humans who generated these algorithms, it can lead to serious harm. Computer programs, for example, have wrongly flagged Black defendants as twice as likely to reoffend as someone who’s white. When an AI used cost as a proxy for health needs, it falsely named Black patients as healthier than equally sick white ones, as less money was spent on them. Even AI used to write a play relied on using harmful stereotypes for casting. 

      Removing sensitive features from the data seems like a viable tweak. But what happens when it’s not enough? 

      Examples of bias in natural language processing are boundless — but MIT scientists have investigated another important, largely underexplored modality: medical images. Using both private and public datasets, the team found that AI can accurately predict self-reported race of patients from medical images alone. Using imaging data of chest X-rays, limb X-rays, chest CT scans, and mammograms, the team trained a deep learning model to identify race as white, Black, or Asian — even though the images themselves contained no explicit mention of the patient’s race. This is a feat even the most seasoned physicians cannot do, and it’s not clear how the model was able to do this. 

      In an attempt to tease out and make sense of the enigmatic “how” of it all, the researchers ran a slew of experiments. To investigate possible mechanisms of race detection, they looked at variables like differences in anatomy, bone density, resolution of images — and many more, and the models still prevailed with high ability to detect race from chest X-rays. “These results were initially confusing, because the members of our research team could not come anywhere close to identifying a good proxy for this task,” says paper co-author Marzyeh Ghassemi, an assistant professor in the MIT Department of Electrical Engineering and Computer Science and the Institute for Medical Engineering and Science (IMES), who is an affiliate of the Computer Science and Artificial Intelligence Laboratory (CSAIL) and of the MIT Jameel Clinic. “Even when you filter medical images past where the images are recognizable as medical images at all, deep models maintain a very high performance. That is concerning because superhuman capacities are generally much more difficult to control, regulate, and prevent from harming people.”

      In a clinical setting, algorithms can help tell us whether a patient is a candidate for chemotherapy, dictate the triage of patients, or decide if a movement to the ICU is necessary. “We think that the algorithms are only looking at vital signs or laboratory tests, but it’s possible they’re also looking at your race, ethnicity, sex, whether you’re incarcerated or not — even if all of that information is hidden,” says paper co-author Leo Anthony Celi, principal research scientist in IMES at MIT and associate professor of medicine at Harvard Medical School. “Just because you have representation of different groups in your algorithms, that doesn’t guarantee it won’t perpetuate or magnify existing disparities and inequities. Feeding the algorithms with more data with representation is not a panacea. This paper should make us pause and truly reconsider whether we are ready to bring AI to the bedside.” 

      The study, “AI recognition of patient race in medical imaging: a modeling study,” was published in Lancet Digital Health on May 11. Celi and Ghassemi wrote the paper alongside 20 other authors in four countries.

      To set up the tests, the scientists first showed that the models were able to predict race across multiple imaging modalities, various datasets, and diverse clinical tasks, as well as across a range of academic centers and patient populations in the United States. They used three large chest X-ray datasets, and tested the model on an unseen subset of the dataset used to train the model and a completely different one. Next, they trained the racial identity detection models for non-chest X-ray images from multiple body locations, including digital radiography, mammography, lateral cervical spine radiographs, and chest CTs to see whether the model’s performance was limited to chest X-rays. 

      The team covered many bases in an attempt to explain the model’s behavior: differences in physical characteristics between different racial groups (body habitus, breast density), disease distribution (previous studies have shown that Black patients have a higher incidence for health issues like cardiac disease), location-specific or tissue specific differences, effects of societal bias and environmental stress, the ability of deep learning systems to detect race when multiple demographic and patient factors were combined, and if specific image regions contributed to recognizing race. 

      What emerged was truly staggering: The ability of the models to predict race from diagnostic labels alone was much lower than the chest X-ray image-based models. 

      For example, the bone density test used images where the thicker part of the bone appeared white, and the thinner part appeared more gray or translucent. Scientists assumed that since Black people generally have higher bone mineral density, the color differences helped the AI models to detect race. To cut that off, they clipped the images with a filter, so the model couldn’t color differences. It turned out that cutting off the color supply didn’t faze the model — it still could accurately predict races. (The “Area Under the Curve” value, meaning the measure of the accuracy of a quantitative diagnostic test, was 0.94–0.96). As such, the learned features of the model appeared to rely on all regions of the image, meaning that controlling this type of algorithmic behavior presents a messy, challenging problem. 

      The scientists acknowledge limited availability of racial identity labels, which caused them to focus on Asian, Black, and white populations, and that their ground truth was a self-reported detail. Other forthcoming work will include potentially looking at isolating different signals before image reconstruction, because, as with bone density experiments, they couldn’t account for residual bone tissue that was on the images. 

      Notably, other work by Ghassemi and Celi led by MIT student Hammaad Adam has found that models can also identify patient self-reported race from clinical notes even when those notes are stripped of explicit indicators of race. Just as in this work, human experts are not able to accurately predict patient race from the same redacted clinical notes.

      “We need to bring social scientists into the picture. Domain experts, which are usually the clinicians, public health practitioners, computer scientists, and engineers are not enough. Health care is a social-cultural problem just as much as it’s a medical problem. We need another group of experts to weigh in and to provide input and feedback on how we design, develop, deploy, and evaluate these algorithms,” says Celi. “We need to also ask the data scientists, before any exploration of the data, are there disparities? Which patient groups are marginalized? What are the drivers of those disparities? Is it access to care? Is it from the subjectivity of the care providers? If we don’t understand that, we won’t have a chance of being able to identify the unintended consequences of the algorithms, and there’s no way we’ll be able to safeguard the algorithms from perpetuating biases.”

      “The fact that algorithms ‘see’ race, as the authors convincingly document, can be dangerous. But an important and related fact is that, when used carefully, algorithms can also work to counter bias,” says Ziad Obermeyer, associate professor at the University of California at Berkeley, whose research focuses on AI applied to health. “In our own work, led by computer scientist Emma Pierson at Cornell, we show that algorithms that learn from patients’ pain experiences can find new sources of knee pain in X-rays that disproportionately affect Black patients — and are disproportionately missed by radiologists. So just like any tool, algorithms can be a force for evil or a force for good — which one depends on us, and the choices we make when we build algorithms.”

      The work is supported, in part, by the National Institutes of Health. More

    • 50 Shares189 Views

      in Data Management & Statistics

      What words can convey

      by

      From search engines to voice assistants, computers are getting better at understanding what we mean. That’s thanks to language-processing programs that make sense of a staggering number of words, without ever being told explicitly what those words mean. Such programs infer meaning instead through statistics — and a new study reveals that this computational approach can assign many kinds of information to a single word, just like the human brain.

      The study, published April 14 in the journal Nature Human Behavior, was co-led by Gabriel Grand, a graduate student in electrical engineering and computer science who is affiliated with MIT’s Computer Science and Artificial Intelligence Laboratory, and Idan Blank PhD ’16, an assistant professor at the University of California at Los Angeles. The work was supervised by McGovern Institute for Brain Research investigator Ev Fedorenko, a cognitive neuroscientist who studies how the human brain uses and understands language, and Francisco Pereira at the National Institute of Mental Health. Fedorenko says the rich knowledge her team was able to find within computational language models demonstrates just how much can be learned about the world through language alone.

      The research team began its analysis of statistics-based language processing models in 2015, when the approach was new. Such models derive meaning by analyzing how often pairs of words co-occur in texts and using those relationships to assess the similarities of words’ meanings. For example, such a program might conclude that “bread” and “apple” are more similar to one another than they are to “notebook,” because “bread” and “apple” are often found in proximity to words like “eat” or “snack,” whereas “notebook” is not.

      The models were clearly good at measuring words’ overall similarity to one another. But most words carry many kinds of information, and their similarities depend on which qualities are being evaluated. “Humans can come up with all these different mental scales to help organize their understanding of words,” explains Grand, a former undergraduate researcher in the Fedorenko lab. For example, he says, “dolphins and alligators might be similar in size, but one is much more dangerous than the other.”

      Grand and Blank, who was then a graduate student at the McGovern Institute, wanted to know whether the models captured that same nuance. And if they did, how was the information organized?

      To learn how the information in such a model stacked up to humans’ understanding of words, the team first asked human volunteers to score words along many different scales: Were the concepts those words conveyed big or small, safe or dangerous, wet or dry? Then, having mapped where people position different words along these scales, they looked to see whether language processing models did the same.

      Grand explains that distributional semantic models use co-occurrence statistics to organize words into a huge, multidimensional matrix. The more similar words are to one another, the closer they are within that space. The dimensions of the space are vast, and there is no inherent meaning built into its structure. “In these word embeddings, there are hundreds of dimensions, and we have no idea what any dimension means,” he says. “We’re really trying to peer into this black box and say, ‘is there structure in here?’”

      Specifically, they asked whether the semantic scales they had asked their volunteers use were represented in the model. So they looked to see where words in the space lined up along vectors defined by the extremes of those scales. Where did dolphins and tigers fall on line from “big” to “small,” for example? And were they closer together along that line than they were on a line representing danger (“safe” to “dangerous”)?

      Across more than 50 sets of world categories and semantic scales, they found that the model had organized words very much like the human volunteers. Dolphins and tigers were judged to be similar in terms of size, but far apart on scales measuring danger or wetness. The model had organized the words in a way that represented many kinds of meaning — and it had done so based entirely on the words’ co-occurrences.

      That, Fedorenko says, tells us something about the power of language. “The fact that we can recover so much of this rich semantic information from just these simple word co-occurrence statistics suggests that this is one very powerful source of learning about things that you may not even have direct perceptual experience with.” More

    • 75 Shares179 Views

      in Data Management & Statistics

      Unlocking new doors to artificial intelligence

      by

      Artificial intelligence research is constantly developing new hypotheses that have the potential to benefit society and industry; however, sometimes these benefits are not fully realized due to a lack of engineering tools. To help bridge this gap, graduate students in the MIT Department of Electrical Engineering and Computer Science’s 6-A Master of Engineering (MEng) Thesis Program work with some of the most innovative companies in the world and collaborate on cutting-edge projects, while contributing to and completing their MEng thesis.

      During a portion of the last year, four 6-A MEng students teamed up and completed an internship with IBM Research’s advanced prototyping team through the MIT-IBM Watson AI Lab on AI projects, often developing web applications to solve a real-world issue or business use cases. Here, the students worked alongside AI engineers, user experience engineers, full-stack researchers, and generalists to accommodate project requests and receive thesis advice, says Lee Martie, IBM research staff member and 6-A manager. The students’ projects ranged from generating synthetic data to allow for privacy-sensitive data analysis to using computer vision to identify actions in video that allows for monitoring human safety and tracking build progress on a construction site.

      “I appreciated all of the expertise from the team and the feedback,” says 6-A graduate Violetta Jusiega ’21, who participated in the program. “I think that working in industry gives the lens of making sure that the project’s needs are satisfied and [provides the opportunity] to ground research and make sure that it is helpful for some use case in the future.”

      Jusiega’s research intersected the fields of computer vision and design to focus on data visualization and user interfaces for the medical field. Working with IBM, she built an application programming interface (API) that let clinicians interact with a medical treatment strategy AI model, which was deployed in the cloud. Her interface provided a medical decision tree, as well as some prescribed treatment plans. After receiving feedback on her design from physicians at a local hospital, Jusiega developed iterations of the API and how the results where displayed, visually, so that it would be user-friendly and understandable for clinicians, who don’t usually code. She says that, “these tools are often not acquired into the field because they lack some of these API principles which become more important in an industry where everything is already very fast paced, so there’s little time to incorporate a new technology.” But this project might eventually allow for industry deployment. “I think this application has a bunch of potential, whether it does get picked up by clinicians or whether it’s simply used in research. It’s very promising and very exciting to see how technology can help us modify, or I can improve, the health-care field to be even more custom-tailored towards patients and giving them the best care possible,” she says.

      Another 6-A graduate student, Spencer Compton, was also considering aiding professionals to make more informed decisions, for use in settings including health care, but he was tackling it from a causal perspective. When given a set of related variables, Compton was investigating if there was a way to determine not just correlation, but the cause-and-effect relationship between them (the direction of the interaction) from the data alone. For this, he and his collaborators from IBM Research and Purdue University turned to a field of math called information theory. With the goal of designing an algorithm to learn complex networks of causal relationships, Compton used ideas relating to entropy, the randomness in a system, to help determine if a causal relationship is present and how variables might be interacting. “When judging an explanation, people often default to Occam’s razor” says Compton. “We’re more inclined to believe a simpler explanation than a more complex one.” In many cases, he says, it seemed to perform well. For instance, they were able to consider variables such as lung cancer, pollution, and X-ray findings. He was pleased that his research allowed him to help create a framework of “entropic causal inference” that could aid in safe and smart decisions in the future, in a satisfying way. “The math is really surprisingly deep, interesting, and complex,” says Compton. “We’re basically asking, ‘when is the simplest explanation correct?’ but as a math question.”

      Determining relationships within data can sometimes require large volumes of it to suss out patterns, but for data that may contain sensitive information, this may not be available. For her master’s work, Ivy Huang worked with IBM Research to generate synthetic tabular data using a natural language processing tool called a transformer model, which can learn and predict future values from past values. Trained on real data, the model can produce new data with similar patterns, properties, and relationships without restrictions like privacy, availability, and access that might come with real data in financial transactions and electronic medical records. Further, she created an API and deployed the model in an IBM cluster, which allowed users increased access to the model and abilities to query it without compromising the original data.

      Working with the advanced prototyping team, MEng candidate Brandon Perez also considered how to gather and investigate data with restrictions, but in his case it was to use computer vision frameworks, centered on an action recognition model, to identify construction site happenings. The team based their work on the Moments in Time dataset, which contains over a million three-second video clips with about 300 attached classification labels, and has performed well during AI training. However, the group needed more construction-based video data. For this, they used YouTube-8M. Perez built a framework for testing and fine-tuning existing object detection models and action recognition models that could plug into an automatic spatial and temporal localization tool — how they would identify and label particular actions in a video timeline. “I was satisfied that I was able to explore what made me curious, and I was grateful for the autonomy that I was given with this project,” says Perez. “I felt like I was always supported, and my mentor was a great support to the project.”

      “The kind of collaborations that we have seen between our MEng students and IBM researchers are exactly what the 6-A MEng Thesis program at MIT is all about,” says Tomas Palacios, professor of electrical engineering and faculty director of the MIT 6-A MEng Thesis program. “For more than 100 years, 6-A has been connecting MIT students with industry to solve together some of the most important problems in the world.” More