Presburger arithmetic, or linear integer arithmetic, is known to have decision procedures that work in triply exponential time.

Jointly with M. Benedikt (Oxford) and A. Mansutti (IMDEA Software Institute), we have recently considered two decidable extensions of Presburger arithmetic: with the power function and with the predicate for the set of powers of 2. No elementary decision procedures were known for these two theories.

In this talk, I will introduce this work and outline ideas behind our results. Namely, we have shown that existence of solutions over N to systems of linear equations and constraints of the form $y = 2^x$ can be decided in nondeterministic exponential time. Also, linear integer arithmetic extended with a predicate for powers of 2 can be decided in triply exponential time.

(Based on a paper in ICALP'23.)

Digitization is the process of creating digital visual twins of real objects and enabling interactions with them in a seamless manner. The human face is a central element of our visual communication and hence its digitization is of utmost importance in visual computing. This has several applications in videoconferencing, cinema production, video games, virtual environments, and many more. In this talk, we will discuss the digitization of the human face. To this end, we will highlight the main stages of the digitization pipeline, and we will outline the main pillars of digitizing the human face. We will show how to achieve these pillars by utilizing the most recent advances in learnable neural implicit scene representations. We will discuss specific face digitization problems in more detail and show important features that learnable neural implicit scene representations can bring to the digitization process. This talk highlights our efforts towards the full digitization of our world, with an ultimate goal of allowing all of us to connect and interact with our friends, family, and loved ones, over a distance.

Concurrent programs are notoriously hard to write correctly, as scheduling nondeterminism introduces subtle errors that are both hard to detect and to reproduce.

Despite rigorous testing, concurrency bugs such as races conditions often find their way into production software, and manifest as critical security issues. Consequently, considerable effort has been made towards developing efficient techniques for detecting them automatically.

The preferred approach to detect data races is through dynamic analysis, where one executes the software with some test inputs, and checks for the presence of bugs in the execution observed.

Traditional bug detectors however are often unable to discover simple bugs present in the underlying software, even after executing the program several times, because these bugs are sensitive to thread scheduling.

In this talk, I will discuss how runtime predictive analysis can help. Runtime predictive analyses aim to expose concurrency bugs, that can be otherwise missed by traditional dynamic analysis techniques (such as the race detector TSan), by inferring the presence of these bugs in alternate executions of the underlying software, without explicitly re-executing the software program.

I will talk about the fundamentals of and recent algorithmic advances for building highly scalable and sound predictive analysis techniques.

After discovering, back in 2011, that Language Models are useful for modeling repetitive patterns in source code (c.f. The "Naturalness" of software ), and exploring some applications thereof, more recently (since about 2019) our group at UC Davis has focused on the observation that Software, as usually written, is bimodal, admitting both the well-known formal, deterministic semantics (mostly for machines) and probabilistic, noisy semantics (for humans). This bimodality property affords both new approaches to software tool construction (using machine-learning) and new ways of studying human code reading. In this talk, I'll give an overview of the Naturalness/Bimodality program, and some recent work we have done on calibrating the quality of code produced by large language models, and also on "bimodal prompting".

--

Please contact the office team for link information

Machine learning algorithms curate much of the content we encounter online. However, there is concern that these algorithms may unintentionally amplify hostile discourse and perpetuate divisive 'us versus them' mentalities. How can we re-engineer algorithms to bridge diverse perspectives and facilitate constructive conflict? First, I will discuss results from our randomized experiment measuring effects of Twitter’s engagement-based ranking algorithm on downstream sociopolitical outcomes like the amplification of divisive content and users’ perceptions of their in-group and out-group. Crucially, we found that an alternative ranking, based on users’ stated preferences rather than their engagement, reduced amplification of negative, partisan, and out-group hostile content. Second, I will discuss how we applied these insights in practice to design an objective function for algorithmic ranking at Twitter. The core idea to the approach is to interpret users' actions in a way that is consistent with their stated, reflective preferences. Finally, I will discuss lessons learned and open questions for designing algorithms that support a plurality of viewpoints, with an emphasis on the emerging paradigm of bridging-based ranking.

In this talk I will give an overview of the role of causality in ethical machine learning, and in particular, in fair and explainable ML. In particular, I will first detail how to use causal reasoning to  study fairness and interpretability problems in algorithmic decision making, stressing the main  limitations that we encounter when aiming to address these problems in practice.  Then, I will provide some hints about how to solve some of these limitations.

With the development of big data, machine learning, and AI, existing software engineering techniques must be re-imagined to provide the productivity gains that developers desire. Furthermore, specialized hardware accelerators like GPUs or FPGAs have become a prominent part of the current computing landscape. However, developing heterogeneous applications is limited to a small subset of programmers with specialized hardware knowledge. To improve productivity and performance for data-intensive and compute-intensive development, now is the time that the software engineering community should design new waves of refactoring, testing, and debugging tools for big data analytics and heterogeneous application development.

In this talk, we overview software development challenges in this new data-intensive scalable computing and heterogeneous computing domain. We describe examples of automated software engineering (debugging, testing, and refactoring) techniques that target this new domain and share lessons learned from building these techniques.

Machine learning provides us with an amazing set of tools to make predictions, but how much should we trust particular predictions? To answer this, we need a way of estimating the confidence we should have in particular predictions of black-box models. Standard tools for doing this give guarantees that are averages over predictions. For instance, in a medical application, such tools might paper over poor performance on one medically relevant demographic group if it is made up for by higher performance on another group. Standard methods also depend on the data distribution being static — in other words, the future should be like the past.

In this talk, I will describe new techniques to address both these problems: a way to produce prediction sets for arbitrary black-box prediction methods that have correct empirical coverage even when the data distribution might change in arbitrary, unanticipated ways and such that we have correct coverage even when we zoom in to focus on demographic groups that can be arbitrary and intersecting. When we just want correct group-wise coverage and are willing to assume that the future will look like the past, our algorithms are especially simple.

This talk is based on two papers, that are joint work with Osbert Bastani, Varun Gupta, Chris Jung, Georgy Noarov, and Ramya Ramalingam.

Please contact the office team for link information

In recent years, humanity has been faced with a new resource - artificial intelligence. AI can be a boon to society, or can also have negative impacts, especially with inappropriate use. My research agenda studies the societal impact of AI, particularly focusing on AI as a resource and on the strategic decisions that agents make in deciding how to use it. In this talk, I will consider some of the key strategic questions that arise in this framework: the decisions that agents make in jointly constructing and sharing AI models, and the decisions that they make in dividing tasks between their own expertise and the expertise of a model. The first of these questions has motivated my work on "model-sharing games", which models scenarios such as federated learning or data cooperatives. In this setting, we view agents with data as game-theoretic players and analyze questions of stability, optimality, and fairness (https://arxiv.org/abs/2010.00753, https://arxiv.org/abs/2106.09580, https:// arxiv.org/abs/2112.00818). Secondly, I will describe some of my ongoing work in modeling human-algorithm collaboration. In particular, I will describe work on best-item recovery in categorical prediction, showing how differential accuracy rates and anchoring on algorithmic suggestions can influence overall performance (https://arxiv.org/abs/2308.11721).

Large language Models (LLMs, e.g., GPT-3, OPT, TNLG,…) are shown to have a remarkably high performance on standard benchmarks, due to their high parameter count, extremely large training datasets, and significant compute. Although the high parameter count in these models leads to more expressiveness, it can also lead to higher memorization, which, coupled with large unvetted, web-scraped datasets can cause different negative societal and ethical impacts such as leakage of private, sensitive information and generation of harmful text. In this talk, we will go over how these issues affect the trustworthiness of LLMs, and zoom in on how we can measure the leakage and memorization of these models, and mitigate it through differentially private training. Finally we will discuss what it would actually mean for LLMs to be privacy preserving, and what are the future research directions on making large models trustworthy.

In this talk, I will explore three elements of designing next-generation machine learning systems: congestion control, network topology, and computation frequency. I will show that fair sharing, the holy grail of congestion control algorithms, is not necessarily desirable for deep neural network training clusters. Then I will introduce a new optical fabric that optimally combines network topology and parallelization strategies for machine learning training clusters. Finally, I will demonstrate the benefits of leveraging photonic computing systems for real-time, energy-efficient inference via analog computing. I will discuss that pushing the frontiers of optical networks for machine learning workloads will enable us to fully harness the potential of deep neural networks and achieve improved performance and scalability.

Computer Vision has become increasingly capable of representing the 3D world, given large sets of dense and complete observations, such as images or 3D scans. However, in comparison with humans, we still lack an important ability: deriving 3D representations from just a few, incomplete 2D observations. Replicating this ability might be the next important step towards a general vision system. A key aspect of the human abilities is that observations are complemented by previously learned information: the world is not only sensed - to a large degree it is inferred. The common way to approach this task with deep learning are data priors, which capture information present in large datasets and which are used to perform inference from novel observations. This talk will discuss 3D data priors and their important connection to 3D representations. Choosing the right representation, we can have abstract control over which information is learned from data and how we can use it during inference, which leads to more effective solutions than simply learning everything end-to-end. Thus, the focus of my research and this talk will be on representations with important properties, such as data efficiency and useful equi- and invariances, which enable the formulation of sophisticated, task-specific data priors. These presented concepts are showcased on examples from my and collaborating groups, e.g., as data priors for reconstructing objects or object interaction sequences from incomplete observations.

Given a sentence $\forall X \exists Y \varphi(X, Y)$ in a first-order theory, it is well-known that there exists a function $F(X)$ for $Y$ in $\varphi$ such that $\exists Y \varphi(X, Y)\leftrightarrow \varphi(X, F(X))$ holds for all values of the universal variables X. Such a function is also called a Skolem function, in honour of Thoralf Skolem who first made us of this in proving what are now known as the Lowenheim-Skolem theorems. The existence of a Skolem function for a given formula is technically analogous to the Axiom of Choice -- it doesn't give us any any hint about how to compute the function, although we know such a function exists. Nevertheless, since Skolem functions are often very useful in practical applications (like finding a strategy for a reactive controller), we investigate when is it possible to algorithmically construct a Turing machine that computes a Skolem function for a given first-order formula. We show that under fairly relaxed conditions, this cannot be done. Does this mean the end of the road for automatic synthesis of Skolem functions? Fortunately, no. We show model-theoretic necessary and sufficient condition for the existence and algorithmic synthesizability of Turing machines implementing Skolem functions. We show that several useful first-order theories satisfy these conditions, and hence admit algorithms that can synthesize Turing machines implementing Skolem functions. We conclude by presenting several open problems in this area.

Language models that have been trained to predict the next word over billions of text documents have been shown to also significantly predict brain recordings of people comprehending language. Understanding the reasons behind the observed similarities between language in machines and language in the brain can lead to more insight into both systems. In this talk, we will discuss a series of recent works that make progress towards this question along different dimensions. The unifying principle among these works that allows us to make scientific claims about why one black box (language model) aligns with another black box (the human brain) is our ability to make specific perturbations in the language model and observe their effect on the alignment with the brain. Building on this approach, these works reveal that the observed alignment is due to more than next-word prediction and word-level semantics and is partially related to joint processing of select linguistic information in both systems. Furthermore, we find that the brain alignment can be improved by training a language model to summarize narratives. Taken together, these works make progress towards determining the sufficient and necessary conditions under which language in machines aligns with language in the brain.

Efficient data-intensive systems translate data into value for decision making. As data is collected at unprecedented rates for timely analysis, the model-centric paradigm of machine learning (ML) is shifting towards a data-centric and system-centric paradigm. Recent breakthroughs in large ML models (e.g., GPT 4 and ChatGPT) and the remarkable outcomes of reinforcement learning (e.g., AlphaFold and AlphaCode) have shown that scalable data management and its optimizations are critical to obtain state-of-the-art performance. This talk aims to answer the question "how to co-design multiple layers of the software/system stack to improve scalability, performance, and energy efficiency of ML and data-intensive systems". It addresses the challenges to build fully automated data-intensive systems that integrate the ML layer, the data management layer, and the compilation-based optimization layer. Finally, this talk will sketch and explore the vision to leverage the computational advantage of quantum computing on hybrid classic/quantum systems in the post-Moore era.

Please contact office for zoom link information.