Today's AI models learn rich internal representations, such as the visual features inside diffusion models that produce generated images, and offer a new kind of material for co-creation. However, interfaces for creating with generative AI typically operate on the level of inputs and outputs, obscuring the material formation that unfolds in between. To address this gap, this pictorial documents a material-driven journey of designing an interface for interacting with the hidden layers of diffusion models.

Distilled diffusion models generate images in far fewer timesteps but suffer from reduced sample diversity when generating multiple outputs from the same prompt. To understand this phenomenon, we first investigate whether distillation damages concept representations by examining if the required diversity is properly learned. Surprisingly, distilled models retain the base model’s representational structure: control mechanisms like Concept Sliders and LoRAs transfer seamlessly without retraining, and Slider-Space analysis reveals distilled models possess variational directions needed for diversity yet fail to activate them. This redirects our investigation to understanding how the generation dynamics differ between base and distilled models. Using ${{\hat{\mathbf x}}_0}$ trajectory visualization, we discover distilled models commit to their final image structure almost immediately at the first timestep, while base models distribute structural decisions across many steps. To test whether this first-step commitment causes the diversity loss, we introduce diversity distillation, a hybrid approach using the base model for only the first critical timestep before switching to the distilled model. This single intervention restores sample diversity while maintaining computational efficiency. We provide both causal validation and theoretical support showing why the very first timestep concentrates the diversity bottleneck in distilled models. Our code and data are available at distillation.baulab.info

A widely used strategy to discover and understand language model mechanisms is circuit analysis.A circuit is a minimal subgraph of a model's computation graph that executes a specific task.We identify a gap in existing circuit discovery methods: they assume circuits are position-invariant, treating model components as equally relevant across input positions.This limits their ability to capture cross-positional interactions or mechanisms that vary across positions.To address this gap, we propose two improvements to incorporate positionality into circuits, even on tasks containing variablelength examples.First, we extend edge attribution patching, a gradient-based method for circuit discovery, to differentiate between token positions.Second, we introduce the concept of a dataset schema, which defines token spans with similar semantics across examples, enabling position-aware circuit discovery in datasets with variable length examples.We additionally develop an automated pipeline for schema generation and application using large language models.Our approach enables fully automated discovery of position-sensitive circuits, yielding better trade-offs between circuit size and faithfulness compared to prior work. 1 Belinkov.2021.Causal analysis of syntactic agreement mechanisms in neural language models.

We present SliderSpace, a framework for automatically decomposing the visual capabilities of diffusion models into controllable and human-understandable directions. Unlike existing control methods that require a user to specify attributes for each edit direction individually, SliderSpace discovers multiple interpretable and diverse directions simultaneously from a single text prompt. Each direction is trained as a low-rank adaptor, enabling compositional control and the discovery of surprising possibilities in the model's latent space. Through extensive experiments on state-of-the-art diffusion models, we demonstrate SliderSpace's effectiveness across three applications: concept decomposition, artistic style exploration, and diversity enhancement. Our quantitative evaluation shows that SliderSpace-discovered directions decompose the visual structure of model's knowledge effectively, offering insights into the latent capabilities encoded within diffusion models. User studies further validate that our method produces more diverse and useful variations compared to baselines. Our code, data and trained weights are available at https://sliderspace.baulab.info

We introduce NNsight and NDIF, technologies that work in tandem to enable scientific study of the representations and computations learned by very large neural networks. NNsight is an open-source system that extends PyTorch to introduce deferred remote execution. The National Deep Inference Fabric (NDIF) is a scalable inference service that executes NNsight requests, allowing users to share GPU resources and pretrained models. These technologies are enabled by the Intervention Graph, an architecture developed to decouple experimental design from model runtime. Together, this framework provides transparent and efficient access to the internals of deep neural networks such as very large language models (LLMs) without imposing the cost or complexity of hosting customized models individually. We conduct a quantitative survey of the machine learning literature that reveals a growing gap in the study of the internals of large-scale AI. We demonstrate the design and use of our framework to address this gap by enabling a range of research methods on huge models. Finally, we conduct benchmarks to compare performance with previous approaches. Code, documentation, and tutorials are available at https://nnsight.net/.

We introduce methods for discovering and applying **sparse feature circuits**. These are causally implicated subnetworks of human-interpretable features for explaining language model behaviors. Circuits identified in prior work consist of polysemantic and difficult-to-interpret units like attention heads or neurons, rendering them unsuitable for many downstream applications. In contrast, sparse feature circuits enable detailed understanding of unanticipated mechanisms in neural networks. Because they are based on fine-grained units, sparse feature circuits are useful for downstream tasks: We introduce SHIFT, where we improve the generalization of a classifier by ablating features that a human judges to be task-irrelevant. Finally, we demonstrate an entirely unsupervised and scalable interpretability pipeline by discovering thousands of sparse feature circuits for automatically discovered model behaviors.

How can we know whether new mechanistic interpretability methods achieve real improvements? In pursuit of lasting evaluation standards, we propose MIB, a Mechanistic Interpretability Benchmark, with two tracks spanning four tasks and five models. MIB favors methods that precisely and concisely recover relevant causal pathways or causal variables in neural language models. The circuit localization track compares methods that locate the model components---and connections between them---most important for performing a task (e.g., attribution patching or information flow routes). The causal variable track compares methods that featurize a hidden vector, e.g., sparse autoencoders (SAE) or distributed alignment search (DAS), and align those features to a task-relevant causal variable. Using MIB, we find that attribution and mask optimization methods perform best on circuit localization. For causal variable localization, we find that the supervised DAS method performs best, while SAEs features are not better than neurons, i.e., non-featurized hidden vectors. These findings illustrate that MIB enables meaningful comparisons, and increases our confidence that there has been real progress in the field.

"Machine unlearning" is a popular proposed solution for mitigating the existence of content in an AI model that is problematic for legal or moral reasons, including privacy, copyright, safety, and more. For example, unlearning is often invoked as a solution for removing the effects of specific information from a generative-AI model's parameters, e.g., a particular individual's personal data or the inclusion of copyrighted content in the model's training data. Unlearning is also proposed as a way to prevent a model from generating targeted types of information in its outputs, e.g., generations that closely resemble a particular individual's data or reflect the concept of "Spiderman." Both of these goals--the targeted removal of information from a model and the targeted suppression of information from a model's outputs--present various technical and substantive challenges. We provide a framework for ML researchers and policymakers to think rigorously about these challenges, identifying several mismatches between the goals of unlearning and feasible implementations. These mismatches explain why unlearning is not a general-purpose solution for circumscribing generative-AI model behavior in service of broader positive impact.

In this work, we introduce Erasure of Language Memory (ELM), a principled approach to concept-level unlearning that operates by matching distributions defined by the model's own introspective classification capabilities. Our key insight is that effective unlearning should leverage the model's ability to evaluate its own knowledge, using the language model itself as a classifier to identify and reduce the likelihood of generating content related to undesired concepts. ELM applies this framework to create targeted low-rank updates that reduce generation probabilities for concept-specific content while preserving the model's broader capabilities. We demonstrate ELM's efficacy on biosecurity, cybersecurity, and literary domain erasure tasks. Comparative evaluation reveals that ELM-modified models achieve near-random performance on assessments targeting erased concepts, while simultaneously preserving generation coherence, maintaining benchmark performance on unrelated tasks, and exhibiting strong robustness to adversarial attacks.

In concept erasure, a model is modified to selectively prevent it from generating a target concept. Despite the rapid development of new methods, it remains unclear how thoroughly these approaches remove the target concept from the model. We begin by proposing two conceptual models for the erasure mechanism in diffusion models: (i) interfering with the model’s internal guidance processes, and (ii) reducing the unconditional likelihood of generating the target concept, potentially removing it entirely. To assess whether a concept has been truly erased from the model, we introduce a comprehensive suite of independent probing techniques: supplying visual context, modifying the diffusion trajectory, applying classifier guidance, and analyzing the model's alternative generations that emerge in place of the erased concept. Our results shed light on the value of exploring concept erasure robustness outside of adversarial text inputs, and emphasize the importance of comprehensive evaluations for erasure in diffusion models.

For large language models (LLMs), sparse autoencoders (SAEs) have been shown to decompose intermediate representations that often are not interpretable directly into sparse sums of interpretable features, facilitating better control and subsequent analysis. However, similar analyses and approaches have been lacking for text-to-image models. We investigate the possibility of using SAEs to learn interpretable features for SDXL Turbo, a few-step text-to-image diffusion model. To this end, we train SAEs on the updates performed by transformer blocks within SDXL Turbo's denoising U-net in its 1-step setting. Interestingly, we find that they generalize to 4-step SDXL Turbo and even to the multi-step SDXL base model (i.e., a different model) without additional training. In addition, we show that their learned features are interpretable, causally influence the generation process, and reveal specialization among the blocks. We do so by creating RIEBench, a representation-based image editing benchmark, for editing images while they are generated by turning on and off individual SAE features. This allows us to track which transformer blocks' features are the most impactful depending on the edit category. Our work is the first investigation of SAEs for interpretability in text-to-image diffusion models and our results establish SAEs as a promising approach for understanding and manipulating the internal mechanisms of text-to-image models.

LLMs are trained to refuse harmful instructions, but do they truly understand harmfulness beyond just refusing? Prior work has shown that LLMs’ refusal behaviors can be mediated by a one-dimensional subspace, i.e., a refusal direction. In this work, we identify a new dimension to analyze safety mechanisms in LLMs, i.e., harmfulness, which is encoded internally as a separate concept from refusal. And there exists a harmfulness direction that is distinct from the refusal direction. As causal evidence, steering along the harmfulness direction can lead LLMs to interpret harmless instructions as harmful, but steering along the refusal direction tends to elicit refusal responses directly without reversing the model’s judgment on harmfulness. Furthermore, using our identified harmfulness concept, we find that certain jailbreak methods work by reducing the refusal signals without suppressing the model’s internal belief of harmfulness. We also find that adversarially fine- tuning models to accept harmful instructions has minimal impact on the model’s internal belief of harmfulness. These insights lead to a practical safety application: The model’s latent harmfulness representation can serve as an intrinsic safeguard (Latent Guard) for detecting unsafe inputs and reducing over-refusals that is robust to finetuning attacks. For instance, our Latent Guard achieves performance comparable to or better than Llama Guard 3 8B, a dedicated finetuned safeguard model, across different jailbreak methods. Our findings suggest that LLMs’ internal understanding of harmfulness is more robust than their refusal decision to diverse input instructions, offering a new perspective to study AI safety.

LLMs process text as sequences of tokens that roughly correspond to words, where less common words are represented by multiple tokens.However, individual tokens are often semantically unrelated to the meanings of the words/concepts they comprise.For example, Llama-2-7b's tokenizer splits the word "northeastern" into the tokens [_n, ort, he, astern], none of which correspond to semantically meaningful units like "north" or "east."Similarly, the overall meanings of named entities like "Neil Young" and multi-word expressions like "break a leg" cannot be directly inferred from their constituent tokens.Mechanistically, how do LLMs convert such arbitrary groups of tokens into useful higher-level representations?In this work, we find that last token representations of named entities and multi-token words exhibit a pronounced "erasure" effect, where information about previous and current tokens is rapidly forgotten in early layers.Using this observation, we propose a method to "read out" the implicit vocabulary of an autoregressive LLM by examining differences in token representations across layers, and present results of this method for Llama-2-7b and Llama-3-8b.To our knowledge, this is the first attempt to probe the implicit vocabulary of an LLM. 1

Much of the knowledge encoded in transformer language models (LMs) may be expressed in terms of relations: relations between words and their synonyms, entities and their attributes, etc. We show that, for a subset of relations, this computation is well-approximated by a single linear transformation on the subject representation. Linear relation representations may be obtained by constructing a first-order approximation to the LM from a single prompt, and they exist for a variety of factual, commonsense, and linguistic relations. However, we also identify many cases in which LM predictions capture relational knowledge accurately, but this knowledge is not linearly encoded in their representations. Our results thus reveal a simple, interpretable, but heterogeneously deployed knowledge representation strategy in transformer LMs.

Fine-tuning on generalized tasks such as instruction following, code generation, and mathematics has been shown to enhance language models' performance on a range of tasks. Nevertheless, explanations of how such fine-tuning influences the internal computations in these models remain elusive. We study how fine-tuning affects the internal mechanisms implemented in language models. As a case study, we explore the property of entity tracking, a crucial facet of language comprehension, where models fine-tuned on mathematics have substantial performance gains. We identify a mechanism that enables entity tracking and show that (i) both the original model and its fine-tuned version implement entity tracking with the same circuit. In fact, the entity tracking circuit of the fine-tuned version performs better than the full original model. (ii) The circuits of all the models implement roughly the same functionality, that is entity tracking is performed by tracking the position of the correct entity in both the original model and its fine-tuned version. (iii) Performance boost in the fine-tuned model is primarily attributed to its improved ability to handle positional information. To uncover these findings, we employ two methods: DCM, which automatically detects model components responsible for specific semantics, and CMAP, a new approach for patching activations across models to reveal improved mechanisms. Our findings suggest that fine-tuning enhances, rather than fundamentally alters, the mechanistic operation of the model.

We report the presence of a simple neural mechanism that represents an input-output function as a vector within autoregressive transformer language models (LMs). Using causal mediation analysis on a diverse range of in-context-learning (ICL) tasks, we find that a small number attention heads transport a compact representation of the demonstrated task, which we call a function vector (FV). FVs are robust to changes in context, i.e., they trigger execution of the task on inputs such as zero-shot and natural text settings that do not resemble the ICL contexts from which they are collected. We test FVs across a range of tasks, models, and layers and find strong causal effects across settings in middle layers. We investigate the internal structure of FVs and find while that they often contain information that encodes the output space of the function, this information alone is not sufficient to reconstruct an FV. Finally, we test semantic vector composition in FVs, and find that to some extent they can be summed to create vectors that trigger new complex tasks. Our findings show that compact, causal internal vector representations of function abstractions can be explicitly extracted from LLMs.

What latent features are encoded in language model (LM) representations? Recent work on training sparse autoencoders (SAEs) to disentangle interpretable features in LM representations has shown significant promise. However, evaluating the quality of these SAEs is difficult because we lack a ground-truth collection of interpretable features which we expect good SAEs to identify. We thus propose to measure progress in interpretable dictionary learning by working in the setting of LMs trained on Chess and Othello transcripts. These settings carry natural collections of interpretable features—for example, “there is a knight on F3”—which we leverage into metrics for SAE quality. To guide progress in interpretable dictionary learning, we introduce a new SAE training technique, $p$-annealing, which demonstrates improved performance on our metric.

Text-to-image models suffer from various safety issues that may limit their suitability for deployment. Previous methods have separately addressed individual issues of bias, copyright, and offensive content in text-to-image models. However, in the real world, all of these issues appear simultaneously in the same model. We present a method that tackles all issues with a single approach. Our method, Unified Concept Editing (UCE), edits the model without training using a closed-form solution, and scales seamlessly to concurrent edits on text-conditional diffusion models.We present scalable simultaneous debiasing, style erasure, and content moderation by editing text-to-image projections, and perform extensive experiments demonstrating improved efficacy and scalability over prior work. Our code is available at unified.baulab.info.

We conjecture that hidden state vectors corresponding to individual input tokens encode information sufficient to accurately predict several tokens ahead.More concretely, in this paper we ask: Given a hidden (internal) representation of a single token at position t in an input, can we reliably anticipate the tokens that will appear at positions ≥ t + 2? To test this, we measure linear approximation and causal intervention methods in GPT-J-6B to evaluate the degree to which individual hidden states in the network contain signal rich enough to predict future hidden states and, ultimately, token outputs.We find that, at some layers, we can approximate a model's output with more than 48% accuracy with respect to its prediction of subsequent tokens through a single hidden state.Finally we present a "Future Lens" visualization that uses these methods to create a new view of transformer states.

Motivated by concerns that large-scale diffusion models can produce undesirable output such as sexually explicit content or copyrighted artistic styles, we study erasure of specific concepts from diffusion model weights. We propose a fine-tuning method that can erase a visual concept from a pre-trained diffusion model, given only the name of the style and using negative guidance as a teacher. We benchmark our method against previous approaches that remove sexually explicit content and demonstrate its effectiveness, performing on par with Safe Latent Diffusion and censored training. To evaluate artistic style removal, we conduct experiments erasing five modern artists from the network and conduct a user study to assess the human perception of the removed styles. Unlike previous methods, our approach can remove concepts from a diffusion model permanently rather than modifying the output at the inference time, so it cannot be circumvented even if a user has access to model weights. Our code, data, and results are available at erasing.baulab.info.

Labeling neural network submodules with human-legible descriptions is useful for many downstream tasks: such descriptions can surface failures, guide interventions, and perhaps even explain important model behaviors. To date, most mechanistic descriptions of trained networks have involved small models, narrowly delimited phenomena, and large amounts of human labor. Labeling all human-interpretable sub-computations in models of increasing size and complexity will almost certainly require tools that can generate and validate descriptions automatically. Recently, techniques that use learned models in-the-loop for labeling have begun to gain traction, but methods for evaluating their efficacy are limited and ad-hoc. How should we validate and compare open-ended labeling tools? This paper introduces FIND (Function INterpretation and Description), a benchmark suite for evaluating the building blocks of automated interpretability methods. FIND contains functions that resemble components of trained neural networks, and accompanying descriptions of the kind we seek to generate. The functions are procedurally constructed across textual and numeric domains, and involve a range of real-world complexities, including noise, composition, approximation, and bias. We evaluate methods that use pretrained language models (LMs) to produce code-based and natural language descriptions of function behavior. Additionally, we introduce a new interactive method in which an Automated Interpretability Agent (AIA) generates function descriptions. We find that an AIA, built with an off-the-shelf LM augmented with black-box access to functions, can sometimes infer function structure—acting as a scientist by forming hypotheses, proposing experiments, and updating descriptions in light of new data. However, FIND also reveals that LM-based descriptions capture global function behavior while missing local details. These results suggest that FIND will be useful for characterizing the performance of more sophisticated interpretability methods before they are applied to real-world models.

The growing proliferation of customized and pretrained generative models has made it infeasible for a user to be fully cognizant of every model in existence. To address this need, we introduce the task of content-based model search: given a query and a large set of generative models, finding the models that best match the query. As each generative model produces a distribution of images, we formulate the search task as an optimization problem to select the model with the highest probability of generating similar content as the query. We introduce a formulation to approximate this probability given the query from different modalities, e.g., image, sketch, and text. Furthermore, we propose a contrastive learning framework for model retrieval, which learns to adapt features for various query modalities. We demonstrate that our method outperforms several baselines on Generative Model Zoo, a new benchmark we create for the model retrieval task.

The CLIP network measures the similarity between natural text and images; in this work, we investigate the entanglement of the representation of word images and natural images in its image encoder. First, we find that the image encoder has an ability to match word images with natural images of scenes described by those words. This is consistent with previous research that suggests that the meaning and the spelling of a word might be entangled deep within the network. On the other hand, we also find that CLIP has a strong ability to match nonsense words, suggesting that processing of letters is separated from processing of their meaning. To explicitly determine whether the spelling capability of CLIP is separable, we devise a procedure for identifying representation subspaces that selectively isolate or eliminate spelling capabilities. We benchmark our methods against a range of retrieval tasks, and we also test them by measuring the appearance of text in CLIP-guided generated images. We find that our methods are able to cleanly separate spelling capabilities of CLIP from the visual processing of natural images.

We analyze the storage and recall of factual associations in autoregressive transformer language models, finding evidence that these associations correspond to localized, directly-editable computations. We first develop a causal intervention for identifying neuron activations that are decisive in a model's factual predictions. This reveals a distinct set of steps in middle-layer feed-forward modules that mediate factual predictions while processing subject tokens. To test our hypothesis that these computations correspond to factual association recall, we modify feed-forward weights to update specific factual associations using Rank-One Model Editing (ROME). We find that ROME is effective on a standard zero-shot relation extraction (zsRE) model-editing task, comparable to existing methods. To perform a more sensitive evaluation, we also evaluate ROME on a new dataset of counterfactual assertions, on which it simultaneously maintains both specificity and generalization, whereas other methods sacrifice one or another. Our results confirm an important role for mid-layer feed-forward modules in storing factual associations and suggest that direct manipulation of computational mechanisms may be a feasible approach for model editing. The code, dataset, visualizations, and an interactive demo notebook are available in the supplemental materials.

Deep generative models make visual content creation more accessible to novice users by automating the synthesis of diverse, realistic content based on a collected dataset. However, the current machine learning approaches miss a key element of the creative process - the ability to synthesize things that go far beyond the data distribution and everyday experience. To begin to address this issue, we enable a user to "warp" a given model by editing just a handful of original model outputs with desired geometric changes. Our method applies a low-rank update to a single model layer to reconstruct edited examples. Furthermore, to combat overfitting, we propose a latent space augmentation method based on style-mixing. Our method allows a user to create a model that synthesizes endless objects with defined geometric changes, enabling the creation of a new generative model without the burden of curating a large-scale dataset. We also demonstrate that edited models can be composed to achieve aggregated effects, and we present an interactive interface to enable users to create new models through composition. Empirical measurements on multiple test cases suggest the advantage of our method against recent GAN fine-tuning methods. Finally, we showcase several applications using the edited models, including latent space interpolation and image editing.

A large body of recent work has identified transformations in the latent spaces of generative adversarial networks (GANs) that consistently and interpretably transform generated images. But existing techniques for identifying these transformations rely on either a fixed vocabulary of prespecified visual concepts, or on unsupervised disentanglement techniques whose alignment with human judgments about perceptual salience is unknown. This paper introduces a new method for building open-ended vocabularies of primitive visual concepts represented in a GAN’s latent space. Our approach is built from three components: (1) automatic identification of perceptually salient directions based on their layer selectivity; (2) human annotation of these directions with free-form, compositional natural language descriptions; and (3) decomposition of these annotations into a visual concept vocabulary, consisting of distilled directions labeled with single words. Experiments show that concepts learned with our approach are reliable and composable—generalizing across classes, contexts, and observers, and enabling fine-grained manipulation of image style and content.

Can a user create a deep generative model by sketching a single example? Traditionally, creating a GAN model has required the collection of a large-scale dataset of exemplars and specialized knowledge in deep learning. In contrast, sketching is possibly the most universally accessible way to convey a visual concept. In this work, we present a method, GAN Sketching, for rewriting GANs with one or more sketches, to make GANs training easier for novice users. In particular, we change the weights of an original GAN model according to user sketches. We encourage the model’s output to match the user sketches through a crossdomain adversarial loss. Furthermore, we explore different regularization methods to preserve the original model’s diversity and image quality. Experiments have shown that our method can mold GANs to match shapes and poses specified by sketches while maintaining realism and diversity. Finally, we demonstrate a few applications of the resulting GAN, including latent space interpolation and image editing.

We propose a methodology for modifying the behavior of a classifier by directly rewriting its prediction rules. Our method requires virtually no additional data collection and can be applied to a variety of settings, including adapting a model to new environments, and modifying it to ignore spurious features.

The Droplet family of tools provides novice programmers with a dual-modality environment to learn, offering both blocks and text representations of code. Our showpiece will highlight the next generation of tools in this family, including the launch of Pencil.io, Amphibian 2.0, and work on the development of Lilypad, the next evolution of Droplet. Lilypad will integrate a frame-based editor, which will facilitate study and deployment of dual-modality instruction in later CS coursework and, potentially, industry applications.

We introduce a simple but effective unsupervised method for generating diverse images. We train a class-conditional GAN model without using manually annotated class labels. Instead, our model is conditional on labels automatically derived from clustering in the discriminator’s feature space. Our clustering step automatically discovers diverse modes, and explicitly requires the generator to cover them. Experiments on standard mode collapse benchmarks show that our method outperforms several competing methods when addressing mode collapse. Our method also performs well on large-scale datasets such as ImageNet and Places365, improving both diversity and standard metrics (e.g., Fréchet Inception Distance), compared to previous methods.

We propose a framework for learning through drawing. Our goal is to learn the correspondence between spoken words and abstract visual attributes, from a dataset of spoken descriptions of images. Building upon recent findings that GAN representations can be manipulated to edit semantic concepts in the generated output, we propose a new method to use such GAN-generated images to train a model using a triplet loss. To apply the method, we develop Audio CLEVRGAN, a new dataset of audio descriptions of GAN-generated CLEVR images, and we describe a training procedure that creates a curriculum of GAN-generated images that focuses training on image pairs that differ in a specific, informative way. Training is done without additional supervision beyond the spoken captions and the GAN. We find that training that takes advantage of GAN-generated edited examples results in improvements in the model's ability to learn attributes compared to previous results. Our proposed learning framework also results in models that can associate spoken words with some abstract visual concepts such as color and size.

Despite the success of Generative Adversarial Networks (GANs), mode collapse remains a serious issue during GAN training. To date, little work has focused on understanding and quantifying which modes have been dropped by a model. In this work, we visualize mode collapse at both the distribution level and the instance level. First, we deploy a semantic segmentation network to compare the distribution of segmented objects in the generated images with the target distribution in the training set. Differences in statistics reveal object classes that are omitted by a GAN. Second, given the identified omitted object classes, we visualize the GAN's omissions directly. In particular, we compare specific differences between individual photos and their approximate inversions by a GAN. To this end, we relax the problem of inversion and solve the tractable problem of inverting a GAN layer instead of the entire generator. Finally, we use this framework to analyze several recent GANs trained on multiple datasets and identify their typical failure cases.

The success of recent deep convolutional neural networks (CNNs) depends on learning hidden representations that can summarize the important factors of variation behind the data. In this work, we describe Network Dissection, a method that interprets networks by providing meaningful labels to their individual units. The proposed method quantifies the interpretability of CNN representations by evaluating the alignment between individual hidden units and visual semantic concepts. By identifying the best alignments, units are given interpretable labels ranging from colors, materials, textures, parts, objects and scenes. The method reveals that deep representations are more transparent and interpretable than they would be under a random equivalently powerful basis. We apply our approach to interpret and compare the latent representations of several network architectures trained to solve a wide range of supervised and self-supervised tasks. We then examine factors affecting the network interpretability such as the number of the training iterations, regularizations, different initialization parameters, as well as networks depth and width. Finally we show that the interpreted units can be used to provide explicit explanations of a given CNN prediction for an image. Our results highlight that interpretability is an important property of deep neural networks that provides new insights into what hierarchical structures can learn.

Despite the recent success of GANs in synthesizing images conditioned on inputs such as a user sketch, text, or semantic labels, manipulating the high-level attributes of an existing natural photograph with GANs is challenging for two reasons. First, it is hard for GANs to precisely reproduce an input image. Second, after manipulation, the newly synthesized pixels often do not fit the original image. In this paper, we address these issues by adapting the image prior learned by GANs to image statistics of an individual image. Our method can accurately reconstruct the input image and synthesize new content, consistent with the appearance of the input image. We demonstrate our interactive system on several semantic image editing tasks, including synthesizing new objects consistent with background, removing unwanted objects, and changing the appearance of an object. Quantitative and qualitative comparisons against several existing methods demonstrate the effectiveness of our method.

There has recently been a surge of work in explanatory artificial intelligence (XAI). This research area tackles the important problem that complex machines and algorithms often cannot provide insights into their behavior and thought processes. XAI allows users and parts of the internal system to be more transparent, providing explanations of their decisions in some level of detail. These explanations are important to ensure algorithmic fairness, identify potential bias/problems in the training data, and to ensure that the algorithms perform as expected. However, explanations produced by these systems is neither standardized nor systematically assessed. In an effort to create best practices and identify open challenges, we describe foundational concepts of explainability and show how they can be used to classify existing literature. We discuss why current approaches to explanatory methods especially for deep neural networks are insufficient. Finally, based on our survey, we conclude with suggested future research directions for explanatory artificial intelligence.

We propose a general framework called Network Dissection for quantifying the interpretability of latent representations of CNNs by evaluating the alignment between individual hidden units and a set of semantic concepts. Given any CNN model, the proposed method draws on a data set of concepts to score the semantics of hidden units at each intermediate convolutional layer. The units with semantics are labeled across a broad range of visual concepts including objects, parts, scenes, textures, materials, and colors. We use the proposed method to test the hypothesis that interpretability is an axis-independent property of the representation space, then we apply the method to compare the latent representations of various networks when trained to solve different classification problems. We further analyze the effect of training iterations, compare networks trained with different initializations, and measure the effect of dropout and batch normalization on the interpretability of deep visual representations. We demonstrate that the proposed method can shed light on characteristics of CNN models and training methods that go beyond measurements of their discriminative power.

Pencil Code is a block-based coding tool that helps beginners work with text-based web programming languages. It has been used to allow help first-time programmers of all ages create programs in JavaScript and CoffeeScript.

This workshop introduces a new educational programming environment Pencil Code, which helps bridge the learning gap between visual programming and text-based coding. The free open-source online tool allows students to work with programs using either blocks or Coffeescript or Javascript text code. Students can switch between visual and text modes at any time.

How do students learn computer science? Successful students climb a ladder of learning and meta-learning. Not only do they learn content, but also they gain proficiency with new processes that aid their learning. Each learner takes her/his own path, of course, but the beginning CS student must clear a number of hurdles that we can identify and help them with. In programming, for example, one might struggle with the leap from visual programming to text programming, or from using well-designed integrated tutorials while learning one language, to the chaos of the Internet when learning another.

The phenomenal growth of video on the Web and the increasing sparseness of meta information associated with it forces us to look for signals from the video content for search/information retrieval and browsing based corpus exploration. A large chunk of users' searching/browsing patterns are centered around people present in the video. Doing it at scale in videos remains hard due to a) the absence of labeled data for such a large set of people and b) the large variation of pose/illumination/expression/age/occlusion/quality etc in the target corpus. We propose a system that can learn and recognize faces by combining signals from large scale weakly labeled text, image, and video corpora. First, consistency learning is proposed to create face models for popular persons. We use the text-image co-occurrence on the web as a weak signal of relevance and learn the set of consistent face models from this very large and noisy training set. Second, efficient and accurate face detection and face tracking is applied. Last, the key faces in each face track is select by clustering to get compact and robust representation. The face tracks are further clustered to get more representative key faces and remove duplicate key faces. For each cluster of face tracks, a combination of majority voting and probabilistic voting is done with the automatically learned models. The effectiveness of our framework is demonstrated by results on image and video corpora, in which we can achieve 92.68% in 37 million images and 80% top-5-precision in 1500 hours videos.