steve abreu

Large language models have transformed AI, yet reliably controlling their outputs remains a challenge. This paper explores activation engineering, where outputs of pre-trained LLMs are controlled by manipulating their activations at inference time. Unlike traditional methods using a single steering vector, we introduce conceptors - mathematical constructs that represent sets of activation vectors as ellipsoidal regions. Conceptors act as soft projection matrices and offer more precise control over complex activation patterns. Our experiments demonstrate that conceptors outperform traditional methods across multiple steering tasks. We further use Boolean operations on conceptors for combined steering goals that empirically outperform additively combining steering vectors on a set of tasks. These results highlight conceptors as a promising tool for more effective steering of LLMs.

In the quest for next-generation sequence modeling architectures, State Space Models (SSMs) have emerged as a potent alternative to transformers, particularly for their computational efficiency and suitability for dynamical systems. This paper investigates the effect of quantization on the S5 model to understand its impact on model performance and to facilitate its deployment to edge and resource-constrained platforms. Using quantization-aware training (QAT) and post-training quantization (PTQ), we systematically evaluate the quantization sensitivity of SSMs across different tasks like dynamical systems modeling, Sequential MNIST (sMNIST) and most of the Long Range Arena (LRA). We present fully quantized S5 models whose test accuracy drops less than 1% on sMNIST and most of the LRA. We find that performance on most tasks degrades significantly for recurrent weights below 8-bit precision, but that other components can be compressed further without significant loss of performance. Our results further show that PTQ only performs well on language-based LRA tasks whereas all others require QAT. Our investigation provides necessary insights for the continued development of efficient and hardware-optimized SSMs.

Modern recurrent layers are emerging as a promising path toward edge deployment of foundation models, especially in the context of large language models (LLMs). Compressing the whole input sequence in a finite-dimensional representation enables recurrent layers to model long-range dependencies while maintaining a constant inference cost for each token and a fixed memory requirement. However, the practical deployment of LLMs in resource-limited environments often requires further model compression, such as quantization and pruning. While these techniques are well-established for attention-based models, their effects on recurrent layers remain underexplored. In this preliminary work, we focus on post-training quantization for recurrent LLMs and show that Mamba models exhibit the same pattern of outlier channels observed in attention-based LLMs. We show that the reason for the difficulty of quantizing SSMs is caused by activation outliers, similar to those observed in transformer-based LLMs. We report baseline results for post-training quantization of Mamba that do not take into account the activation outliers and suggest first steps for outlier-aware quantization.

Spiking neural networks and neuromorphic hardware platforms that emulate neural dynamics are slowly gaining momentum and entering main-stream usage. Despite a well-established mathematical foundation for neural dynamics, the implementation details vary greatly across different platforms. Correspondingly, there are a plethora of software and hardware implementations with their own unique technology stacks. Consequently, neuromorphic systems typically diverge from the expected computational model, which challenges the reproducibility and reliability across platforms. Additionally, most neuromorphic hardware is limited by its access via a single software frameworks with a limited set of training procedures. Here, we establish a common reference-frame for computations in neuromorphic systems, dubbed the Neuromorphic Intermediate Representation (NIR). NIR defines a set of computational primitives as idealized continuous-time hybrid systems that can be composed into graphs and mapped to and from various neuromorphic technology stacks. By abstracting away assumptions around discretization and hardware constraints, NIR faithfully captures the fundamental computation, while simultaneously exposing the exact differences between the evaluated implementation and the idealized mathematical formalism. We reproduce three NIR graphs across 7 neuromorphic simulators and 4 hardware platforms, demonstrating support for an unprecedented number of neuromorphic systems. With NIR, we decouple the evolution of neuromorphic hardware and software, ultimately increasing the interoperability between platforms and improving accessibility to neuromorphic technologies. We believe that NIR is an important step towards the continued study of brain-inspired hardware and bottom-up approaches aimed at an improved understanding of the computational underpinnings of nervous systems.

Intelligent assistance involves not only understanding but also action. Existing ego-centric video datasets contain rich annotations of the videos, but not of actions that an intelligent assistant could perform in the moment. To address this gap, we release PARSE-Ego4D, a new set of personal action recommendation annotations for the Ego4D dataset. We take a multi-stage approach to generating and evaluating these annotations. First, we used a prompt-engineered large language model (LLM) to generate context-aware action suggestions and identified over 18,000 action suggestions. While these synthetic action suggestions are valuable, the inherent limitations of LLMs necessitate human evaluation. To ensure high-quality and user-centered recommendations, we conducted a large-scale human annotation study that provides grounding in human preferences for all of PARSE-Ego4D. We analyze the inter-rater agreement and evaluate subjective preferences of participants. Based on our synthetic dataset and complete human annotations, we propose several new tasks for action suggestions based on ego-centric videos. We encourage novel solutions that improve latency and energy requirements. The annotations in PARSE-Ego4D will support researchers and developers who are working on building action recommendation systems for augmented and virtual reality systems.

We provide a perspective on the fundamental relationship between physics and computation, exploring the conditions under which a physical system can be harnessed for computation and the practical means to achieve this. Unlike traditional digital computers that impose discreteness on continuous substrates, unconventional computing embraces the inherent properties of physical systems. Exploring simultaneously the intricacies of physical implementations and applied computational paradigms, we discuss the interdisciplinary developments of unconventional computing. Here, we focus on the potential of photonic substrates for unconventional computing, implementing artificial neural networks to solve data-driven machine learning tasks. Several photonic neural network implementations are discussed, highlighting their potential advantages over electronic counterparts in terms of speed and energy efficiency. Finally, we address the challenges of achieving learning and programmability within physical substrates, outlining key strategies for future research.

The value of brain-inspired neuromorphic computers critically depends on our ability to program them for relevant tasks. Currently, neuromorphic hardware often relies on machine learning methods adapted from deep learning. However, neuromorphic computers have potential far beyond deep learning if we can only harness their energy efficiency and full computational power. Neuromorphic programming will necessarily be different from conventional programming, requiring a paradigm shift in how we think about programming. This paper presents a conceptual analysis of programming within the context of neuromorphic computing, challenging conventional paradigms and proposing a framework that aligns more closely with the physical intricacies of these systems. Our analysis revolves around five characteristics that are fundamental to neuromorphic programming and provides a basis for comparison to contemporary programming methods and languages. By studying past approaches, we contribute a framework that advocates for underutilized techniques and calls for richer abstractions to effectively instrument the new hardware class.

Imaging flow cytometry systems play a critical role in the identification and characterization of large populations of cells or micro-particles. Such systems typically leverage deep artificial neural networks to classify samples. Here we show that an event-based camera and neuromorphic processor can be used in a flow cytometry setup to solve a binary particle classification task with less memory usage, and promising improvements in latency and energy scaling. To reduce the complexity of the spiking neural network, we combine the event-based camera with a free-space optical setup which acts as a non-linear high-dimensional feature map that is computed at the speed of light before the event-based camera receives the signal. We demonstrate, for the first time, a spiking neural network running on neuromorphic hardware for a fully event-based flow cytometry pipeline with 98.45 testing accuracy. Our best artificial neural network on frames of the same data reaches only 97.51, establishing a neuromorphic advantage also in classification accuracy. We further show that our system will scale favorably to more complex classification tasks. We pave the way for real-time classification with throughput of up to 1,000 samples per second and open up new possibilities for online and on-chip learning in flow cytometry applications.

This manuscript serves a specific purpose: to give readers from fields such as material science, chemistry, or electronics an overview of implementing a reservoir computing (RC) experiment with her/his material system. Introductory literature on the topic is rare and the vast majority of reviews puts forth the basics of RC taking for granted concepts that may be nontrivial to someone unfamiliar with the machine learning field. This is unfortunate considering the large pool of material systems that show nonlinear behavior and short-term memory that may be harnessed to design novel computational paradigms. RC offers a framework for computing with material systems that circumvents typical problems that arise when implementing traditional, fully fledged feedforward neural networks on hardware, such as minimal device-to-device variability and control over each unit/neuron and connection. Instead, one can use a random, untrained reservoir where only the output layer is optimized, for example, with linear regression. In the following, we will highlight the potential of RC for hardware-based neural networks, the advantages over more traditional approaches, and the obstacles to overcome for their implementation. Preparing a high-dimensional nonlinear system as a well-performing reservoir for a specific task is not as easy as it seems at first sight. We hope this tutorial will lower the barrier for scientists attempting to exploit their nonlinear systems for computational tasks typically carried out in the fields of machine learning and artificial intelligence. A simulation tool to accompany this paper is available online.

Machine learning has made tremendous progress in recent years and received large amounts of public attention. Though we are still far from designing a full artificially intelligent agent, machine learning has brought us many applications in which computers solve human learning tasks remarkably well. Much of this progress comes from a recent trend within machine learning, called deep learning. Deep learning models are responsible for many state-of-the-art applications of machine learning. Despite their success, deep learning models are hard to train, very difficult to understand, and often times so complex that training is only possible on very large GPU clusters. Lots of work has been done on enabling neural networks to learn efficiently. However, the design and architecture of such neural networks is often done manually through trial and error and expert knowledge. This thesis inspects different approaches, existing and novel, to automate the design of deep feedforward neural networks in an attempt to create less complex models with good performance that take away the burden of deciding on an architecture and make it more efficient to design and train such deep networks.