IRC-Safe Graph Autoencoder for Unsupervised Anomaly Detection

1. Introduction

New physics searches at the high-energy frontier of the Large Hadron Collider (LHC) have so far not resulted in any significant deviation of experimental results from the Standard Model (SM) expectation. However, with a growing dataset of these high energy measurements, the pressure mounts for theoretically motivated scenarios of beyond the SM (BSM) physics which have been devised to tackle known shortcomings of the SM. So what are the ways out of this juxtaposition of experimental agreement with the SM and its failure to describe established physics at small and large distances?

On the one hand, there is an increasing emphasis on theoretically as-model-independent-as-possible approaches based on effective field theory (EFT) (Weinberg, 1979). EFT navigates QFT correlations away from the SM prediction in any possible direction given the SM symmetry and particle content, thus avoiding UV model biases. Alas, such an approach poses its own challenges: looking for deviations from the SM expectation along these lines involves many ad-hoc interactions. Concrete models will typically only source a subset of relevant interactions, e.g., (Englert et al., 2019; Das Bakshi et al., 2021; Bakshi et al., 2022). There has been great progress to facilitate matching calculations (Carmona et al., 2021), however, depending on the new physics scenario, this can create a significant overhead that must be included in the parameter fitting procedure itself (Freitas et al., 2016; Englert et al., 2020).

On the other hand, we can look for phenomenological deviations from specific SM signatures directly in collider results without any new physics bias. Under the assumption that collider data can be modeled sufficiently adequately, we can employ the SM expectation to identify regions where measurements do not follow the SM expectation. This anomaly detection has emerged as a powerful tool to look for any hidden signature of new physics in the data. Recently, a range of state-of-the-art methods for anomaly detection (Collins et al., 2018, 2021; Aaboud et al., 2019; Blance et al., 2019; De Simone and Jacques, 2019; Blance and Spannowsky, 2020; Cheng et al., 2020; Hajer et al., 2020; Nachman, 2020; Nachman and Shih, 2020; Araz and Spannowsky, 2021; Atkinson et al., 2021b; Hallin et al., 2021; Canelli et al., 2022) using deep learning have been designed.

Theoretical consistency when confronting collider data with theoretical expectations is pivotal. The formulation of infrared and collinear (IRC)-safe observables is necessary to guarantee the comparability of experimental measurements and theoretical predictions to all orders in perturbation theory employing the Kinoshita-Lee-Nauenberg (KLN) theorem (Kinoshita, 1962; Lee and Nauenberg, 1964) and collinear factorization (Collins et al., 1989) of parton distributions. Any sensitivity enhancement observed by algorithms that inadvertently employ IRC-unsafe information will be critically assessed in subsequent studies, with a potentially meaningless theoretical outcome. This is a tedious task all too familiar from the use of IRC-unsafe jet clustering algorithms (e.g., iterative cone algorithms) by the CDF and D0 experiments (Kilgore and Giele, 1997), which was later only partially addressed with the midpoint algorithm during Tevatron Run II, until fully IRC-safe algorithms (Catani et al., 1993; Cacciari et al., 2008) were established as the only theoretically meaningful community consensus.

It seems prudent to avoid mistakes of the past: in this paper, we devise an IRC-safe Graph Neural Network (GNN) autoencoder algorithm, employing an Energy-Weighted Message-Passing Network (EMPN) (Konar et al., 2022) for unsupervised anomaly detection. While the IRC-safe loss function is the primary observable of our autoencoder, we also study the latent space (graph) representation structure as a motivating tool for new physics discrimination (Atkinson et al., 2021b; Dillon et al., 2021) and highlight the relations to known and more “traditional” IRC-safe observables. This paper is organized as follows: in Section 2, we outline our EMPN approach and detail our IRC-safe graph construction before we introduce the IRC-safe autoencoder architecture and simulation framework in Section 3. Section 4 is devoted to the discussion of the sensitivity performance of the autoencoder; we also highlight the correlation of sensitivity with more traditional jet-based observables. We conclude in Section 5.

2. A Brief Outline of Energy-Weighted Message Passing Algorithm

This section presents a brief overview of the IRC safe Energy-weighted Message passing algorithm (Konar et al., 2022). It generalizes Energy Flow Networks (Komiske et al., 2019; Dolan and Ore, 2021), an IRC safe feature extraction on point clouds, by learning relational information between two elements (nodes) by constructing a graph out of the point cloud. This procedure is similar to message-passing networks like the Dynamic Graph Convolutional Neural Network (DGCNN) (Wang et al., 2019) that extract local features beyond the global feature extraction via point-cloud-based architectures such as deep-sets (Zaheer et al., 2017) and PointNet (Charles et al., 2017; Qi et al., 2017). The algorithm consists of two necessary ingredients:

• An IRC safe prescription for constructing graphs which guarantees that the graph is invariant under soft and collinear splittings;

• An energy-weighted summed aggregation of messages (and node features after the final message-passing layer) taking the directional (unit vectors or angles) inputs ${\widehat{p}}_i$ and ${\widehat{p}}_j$ of the nodes connected by an edge (j, i) at the initial layer.

In the following, we discuss these two elements separately.

2.1. IRC Safe Graph Construction

The inductive biases that a message-passing algorithm imposes on its input data are highly dependent on the graph structure. For instance, the neighborhood sets (the set of particles a node is connected with) determine the local connectivity of the nodes. Thus, the graph construction algorithm from a point cloud gives a strong indication that a graph neural network is the best avenue to pursue. The IRC safety of a message-passing algorithm also depends on the graph construction, and we highlight such an IRC safe graph construction algorithm in this section.

Let ${S}=\left\{p_1,p_2,p_3,.....,p_N\right\}$ be the set of four-vectors of the particles within a jet, while ${{S}}^{\prime }=\left\{p_1,p_2,....,p_{N+1}\right\}$ is the same set in the presence of an additional splitting. The collinear limit is when the emitted particles r and s with the angular separation Δ_rs tending to zero, while the soft limit refers to the case when one of the particle's energy tends to zero. These four-vectors can be written as

$\begin{array}{ll}{p}_i=(z_i,{\widehat{p}}_i),\text{ with }{z}_i=\frac{p_T^i}{{\displaystyle {\sum }_{j\in {S}}}{p}_T^j}\text{ and }\widehat{p}=(\eta ,\varphi ),& (1)\end{array}$

for hadron colliders, with the separation in the η − ϕ plane between two particles i and j, defined as $\Delta R_{ij}=\sqrt{\Delta {\eta }_{ij}^2+\Delta {\varphi }_{ij}^2}$ denoting the quantity analogous to Δ_ij. Since we will be taking directed edges, the neighborhood set of a node i will be the set of all nodes with incoming connections to i. For all particles i in ${S}$ or ${{S}}^{\prime }$, a graph construction algorithm will construct neighborhood sets ${N}\left[i\right]$ and ${{N}}^{\prime }\left[i\right]$, respectively. We will use a “closed” neighborhood with $i\in {N}\left[i\right]$ instead of an “open” neighborhood $i\notin {N}(i)$, since the second choice will always be IRC unsafe when the node i splits. To illustrate this, we show the radius graph with R₀ in the (η, ϕ) plane in Figure 1, where the node q undergoes a splitting. The black arrows highlight the connections of the radius graph. Figure 1 also demonstrates a nearest neighborhood connection as an example of an IRC unsafe graph construction.

FIGURE 1

Figure 1. Representation of radius graph with R₀ in the (η, ϕ) plane undergoing a QCD splitting. The black arrows correspond to the connections of a radius graph, while the red arrows highlight the 3-nearest neighbors connections. One can see that the radius neighborhoods have the same total energy, which is not the case for those obtained by the nearest neighbors method, leading to an IRC-unsafe construction.

To formalize the graph construction algorithm in terms of the four-vectors of the particles, we define a decision function D(p_i, p_j) and a threshold function T(p_i, p_j), such that any particle j with four-vector p_j will be assigned to the neighborhood of particle i with four-vector p_i if D(p_i, p_j) is less-than or equal-to T(p_i, p_j). This can be summarized as

$\begin{array}{ll}\text{D}(p_i,p_j)\le \text{T}(p_i,p_j)\Rightarrow j\in {N}\left[i\right].& (2)\end{array}$

Since we are interested in the soft and collinear limits, constructing an IRC safe graph requires putting conditions on these functions in the respective kinematical configurations.

The required condition on these functions for a “parent” splitting q → r + s when the “daughters” r, s become collinear is

$\begin{array}{r}\text{D}(p_i,p_r+p_s)\le \text{T}(p_i,p_r+p_s)\iff \text{D}(p_i,p_r)\le \text{T}(p_i,p_r)\wedge \\ \text{D}(p_i,p_s)\le \text{T}(p_i,p_s),\\ \text{D}(p_r+p_s,p_i)\le \text{T}(p_r+p_s,p_i)\iff \text{D}(p_r,p_i)\le \text{T}(p_r,p_i)\wedge \\ \text{D}(p_s,p_i)\le \text{T}(p_s,p_i),& (3)\end{array}$

where the second condition arises since the nodes q, r or s can also be the node whose neighborhood is being determined. The only requirement in the IR limit for a daughter particle is that all the particles in the set ${N}\left[i\right]$ are also present in ${{N}}^{\prime }\left[i\right]$, with the only potential addition of a soft particle. This is guaranteed by the form of Equation (2), since both functions depend only on the four-vector of the two nodes of interest¹. The conditions (c.f. Equation 3) are satisfied in the collinear limit Δ_rs → 0 if

$\begin{array}{ll}\text{D}=\text{D}({\widehat{p}}_i,{\widehat{p}}_j),\text{T}=\text{T}({\widehat{p}}_i,{\widehat{p}}_j),& (4)\end{array}$

employing the definitions Equation (1). Therefore, graphs formed by connecting particles within a constant radius R₀ in the η − ϕ plane are IRC safe when the decision and threshold functions take the form

$\begin{array}{ll}\text{D}=\Delta R_{ij},\text{T}=R_0.& (5)\end{array}$

Note that these choices of D, T yield closed neighborhoods without additional requirements. We will use these graphs in the remainder of this paper; the neighborhood of a particle of such a radius graph is shown in Figure 1.

2.2. Energy-Weighted Message Passing

We detail the IRC safe message passing operation in this section. Before doing so, we summarize the general definition of message passing operation in the following steps. The first step, the message-passing stage, involves calculating the messages for all edges present in the graph. The message function, parameterized as a multilayer perceptron shared for all edges, takes the node features of the two nodes connected by an edge and evaluates the message. Since the message function does not need to be symmetric for the two node features, a direction convention is necessary for the second phase. In our convention, the message originates from all nodes in the neighborhood ${N}\left[i\right]$ and flows toward the particle i. The second step, the node-readout stage, updates the node features of each node in the graph as a permutation-invariant function of all incoming messages.

IRC safety of the updated node features after a message-passing operation is crucially dependent on the nature of the node readout. A readout based on the maximum or minimum value of the node features depends on a single node feature in the neighborhood, and a soft or collinear splitting of this particular node would render the updated node feature IRC-unsafe. This is ultimately related to identifying a specific node in the neighborhood as special², which impedes KLN cancellations. A mean readout, on the other hand, explicitly depends on the cardinality of the neighborhood sets ${N}\left[i\right]$ which is not a well-defined QCD quantity either since there can be an arbitrary but finite amount of resolvable emissions in the enhanced collinear or soft regions of phase space. Thus we use a summed readout, which will inclusively take all the particles in the neighborhood into account and will not explicitly depend on their size.

An IRC safe graph construction algorithm ensures two things: the equality of the sum of energy (transverse energy in the case of hadron colliders) of all particles in either neighborhood sets and the presence of both collinear daughters in ${{N}}^{\prime }\left[i\right]$ if the parent is present in ${N}\left[i\right]$. Defining a scope-dependent energy weight-factors analogous to z_i as

$\begin{array}{l}{\omega }_j^{({K})}=\frac{p_T^j}{{\displaystyle {\sum }_{k\in {K}}}{p}_T^k},\end{array}$

with ${K}$ denoting the set of particles in the particular readout operation, any message passing of the form

$\begin{array}{ll}{\text{h}}_i^{(l+1)}={\displaystyle \sum _{i\in {N}\left[i\right]}}{\omega }_j^{({N}\left[i\right])}{\widehat{\Phi }}^{(l)}({\text{h}}_i^{(l)},{\text{h}}_j^{(l)}),& (6)\end{array}$

with ${\text{h}}_i^{(0)}={\widehat{p}}_i$ and ${\text{h}}_i^{(l)}$ denoting the updated node-features after l message-passing operations satisfies IRC safety; in the infrared limit, it is straightforward to see that any soft particle with $z_r\to 0\Rightarrow {\omega }_r^{({N}\left[i\right])}\to 0$ for any node i. The splitting q → r + s for IRC-safe graphs therefore yields

$\begin{array}{ll}{\omega }_q^{({N}\left[i\right])}={\omega }_r^{({N}\left[i\right])}+{\omega }_s^{({N}\left[i\right])}.& (7)\end{array}$

In the collinear limit with ${\widehat{p}}_q={\widehat{p}}_r={\widehat{p}}_s$ we have ${\widehat{\Phi }}^{(0)}({\widehat{p}}_i,{\widehat{p}}_q)={\widehat{\Phi }}^{(0)}({\widehat{p}}_i,{\widehat{p}}_r)={\widehat{\Phi }}^{(0)}({\widehat{p}}_i,{\widehat{p}}_s)$. Combining this with Equation (7), we obtain (for l = 0)

$\begin{array}{l}{\omega }_q^{({N}\left[i\right])}{\widehat{\Phi }}^{(0)}({\widehat{p}}_i,{\widehat{p}}_q)={\omega }_r^{({N}\left[i\right])}{\widehat{\Phi }}^{(0)}({\widehat{p}}_i,{\widehat{p}}_r)+{\omega }_s^{({N}\left[i\right])}{\widehat{\Phi }}^{(0)}({\widehat{p}}_i,{\widehat{p}}_s).\end{array}$

When evaluating Equation (6) for the neighborhood of a node i, the terms on the RHS and LHS of this expression are the only ones which will not be common between ${N}\left[i\right]$ and ${{N}}^{\prime }\left[i\right]$, due to the IRC safe graph construction. The same expression is followed when i = q on the left, and i = r or i = s on the right, since $\left\{r,s\right\}\subset {{N}}^{\prime }\left[s\right]$ and $\left\{r,s\right\}\subset {{N}}^{\prime }\left[r\right]$, with all three neighborhoods (including ${N}\left[q\right]$) containing the same particles except for q, r, and s. Therefore, from Equation (6), we have ${\text{h}}_q^{(1)}={\text{h}}_r^{(1)}={\text{h}}_s^{(1)}$ for collinear splittings. On the other hand, for a soft daughter, say r, we have ${\text{h}}_q^{(1)}={\text{h}}_s^{(1)}$, but ${\text{h}}_r^{(1)}\ne {\text{h}}_q^{(1)}$, with ${\text{h}}_r^{(1)}$ not necessarily zero. The presence of the node features of the daughter particles, even in the soft or collinear limit, impedes an IRC safe examination of the full jet graph unless observables are specifically designed to be insensitive to their presence in the IRC limit. The procedures to take care of these additional nodes are explained in the following sections, which are different for supervised and unsupervised methods. Since the above derivation used the collinearity of q, r, and s, for IRC safe neighborhoods, for the same neighborhoods and any successive application of an energy-weighted message passing of the form Equation (6), we have ${\text{h}}_q^{(l)}={\text{h}}_r^{(l)}={\text{h}}_s^{(l)}$ for any l.

3. IRC-Safe Graph Autoencoder

In a supervised machine learning scenario, the IRC-safe graph readout acting on the node features of the final message-passing operation gives an IRC-safe graph representation, and one loses the graph's structure. The graph representation, a fixed-length vector obtained after applying a permutation invariant function on the node features for any variable-length graph, feeds into the downstream network. Therefore, training a classifier on the loss function defined with the downstream network's output proceeds without any complications from the presence of additional soft or collinear nodes. On the other hand, a graph autoencoder similar to the one proposed in Atkinson et al. (2021a) preserves the graph structure until the output. Therefore, the autoencoder's output graph will have additional nodes in the soft and collinear limits in the case of extra emissions. Since the observable of interest for anomaly detection with an autoencoder is the loss function, we need to ensure its IRC safety. In this section, we first devise an IRC safe loss function and give details of the network architecture and training.

3.1. An IRC-Safe Loss Function

The definition of the loss function involves input which changes with a soft or collinear splitting. Therefore, the loss which is normally used as an observable in anomaly detection, needs to be IRC-safe. A simple IRC-safe loss function for a jet with constituent set ${G}$ is of the form

$\begin{array}{ll}{{L}}_{{G}}={\displaystyle \sum _{i\in {G}}}{z}_{i}d({\widehat{p}}_i,{\widehat{\stackrel{-}{p}}}_i).& (8)\end{array}$

The barred quantities are the output of the network, while the unbarred quantities are the inputs to the network. The function $d({\widehat{p}}_i,{\widehat{\stackrel{-}{p}}}_i)\ge d_0$ denotes a well-behaved metric (one-to-one) between the input and the output space, with $d({\widehat{p}}_i,{\widehat{p}}_i)=d_0$. We now show that this is indeed an IRC safe choice:

Any soft particle s, will not contribute to the sum since z_s → 0, and hence it is IR safe. For the splitting q → r + s we have

$\begin{array}{l}{{L}}_{{S}}=...+z_{q}d({\widehat{p}}_q,{\widehat{\stackrel{-}{p}}}_q)+...\\ {{L}}_{{{S}}^{\prime }}=...+z_{r}d({\widehat{p}}_r,{\widehat{\stackrel{-}{p}}}_r)+z_{s}d({\widehat{p}}_s,{\widehat{\stackrel{-}{p}}}_s)+....\end{array}$

Since, by construction, a GNN's node output after L total message-passing operations ${\text{h}}_i^{(L)}={\widehat{\stackrel{-}{p}}}_i$, is a function of the input four-vectors {p₁, p₂, p₃, ....p_N}, in general, they can have a very complicated dependence on all the input node features. However, due to the IRC safety of the EMPN, we have

$\begin{array}{ll}{{L}}_{{{S}}^{\prime }}-{{L}}_{{S}}=z_{r}d({\widehat{p}}_r,{\widehat{\stackrel{-}{p}}}_r)+z_{s}d({\widehat{p}}_s,{\widehat{\stackrel{-}{p}}}_s)-z_{q}d({\widehat{p}}_q,{\widehat{\stackrel{-}{p}}}_q).& (9)\end{array}$

In the collinear limit with ${\widehat{p}}_q={\widehat{p}}_r={\widehat{p}}_s\Rightarrow {\widehat{\stackrel{-}{p}}}_q={\widehat{\stackrel{-}{p}}}_r={\widehat{\stackrel{-}{p}}}_s$, we therefore have (since z_q = z_r + z_s),

$\begin{array}{ll}{z}_{q}d({\widehat{p}}_q,{\widehat{\stackrel{-}{p}}}_q)=z_{r}d({\widehat{p}}_r,{\widehat{\stackrel{-}{p}}}_r)+z_{s}d({\widehat{p}}_s,{\widehat{\stackrel{-}{p}}}_s)\Rightarrow {{L}}_{{{S}}^{\text{′}}}-{{L}}_{{S}}=0,& (10)\end{array}$

i.e., collinear safety. In the following analysis of the EMPN autoencoder we will use mean-squared error between the input and output node features for $d({\widehat{p}}_i,{\widehat{\stackrel{-}{p}}}_i)$.

3.2. Jet Graph Definition

To demonstrate the performance of the described algorithm, we use the publicly available top-tagging dataset of Butter et al. (2018) and Kasieczka et al. (2019). The dataset contains a training, validation and testing set of 600k, 200k, and 200k QCD jets, respectively. The training and validation are done only with the background QCD samples since the total cross-section of their production would be orders of magnitude higher than most probable signals. Although the dataset has the same number of top jets for each of the three analysis stages, we use the 200k top jets of the test dataset as a benchmark signal scenario. These jets are simulated using Pythia8 (Sjostrand et al., 2008; Sjöstrand et al., 2015) and passed through Delphes3 (de Favereau et al., 2014) for the detector simulations using the default ATLAS parameter card. Jets are clustered from particle flow (Eflow) constituents with a distance parameter ΔR = 0.8 using the anti-k_t algorithm (Cacciari et al., 2008). The transverse momentum of the jets is in the range p_T ∈ [550, 650] GeV.

Using the constituents of these jets, we construct the radius graphs which serve as the input to the IRC safe graph network. To construct the jet radius graph, we first calculate the inter-particle distance ΔR_ij in the (η, ϕ) plane. Next, we define a set of all the particles i as the neighborhood set ${N}\left[i\right]$ such that ΔR_ij ≤ R₀, where R₀ is an external tunable parameter. Each node is associated with three node features

$\begin{array}{ll}{\text{h}}_i^0=(\Delta {\eta }_i,\Delta {\varphi }_i,\Delta R_i),& (11)\end{array}$

where Δη_i, Δϕ_i, ΔR_i are calculated with respect to the jet axis. For the network analysis, we choose R₀ = 0.3. Since the dependence of the classification power on R₀ for the supervised case was found to be mild (Konar et al., 2022), with the AUC values changing in the third decimal value for different values of R₀ between 0.1 and 0.5, we restrict ourselves to a single value in the intermediate range. The final node vectors contain information about the L-hop neighborhood with an effective radius of R₀ × L. On the other hand, the primary region of activity for the one-prong QCD jets used to train the network lies in a relatively small central region of the total jet of radius ΔR = 0.8. Therefore, the features learnt by the autoencoder would be weakly dependent on R₀, once the effective radius covers a significant portion of the central region.

3.3. Network Architecture and Training

Now that we have described the construction of the jet graphs, we discuss the details of the network architecture and training in this section. Follow from Figure 2 where we sketch a schematic diagram of an IRC safe graph-autoencoder. The encoder consists of three edge convolution operations with output dimensions of 128, 64 and 2, which is the dimension of the latent representation. Since we take three-dimensional node features, we restrict ourselves to a 2-dimensional latent space (g₁, g₂) to induce an information bottleneck³. The decoder also has three edge convolution operations, with the first two dimensions mirroring the encoder network dimensions (excluding the latent dimension). Finally, the last edge convolution operation maps the 128-dimensional node vectors at the penultimate message passing the layer to a three-dimensional space to reconstruct the input node features.

FIGURE 2

Figure 2. A schematic diagram of an IRC safe graph-autoencoder.

We take ${\widehat{\Phi }}^{(l)}$ at each message-passing layer to be a multilayer perceptron (MLP). For an edge convolution operation, we have for two node features ${\text{h}}_i^{(l)}$ and ${\text{h}}_i^{(l)}$ connected by an edge in Equation (6),

$\begin{array}{l}{\widehat{\Phi }}^{(l)}({\text{h}}_i^{(l)},{\text{h}}_j^{(l)})={\widehat{\Phi }}^{(l)}\left({\text{h}}_i^{(l)}\oplus [{\text{h}}_j^{(l)}-{\text{h}}_i^{(l)}]\right).\end{array}$

Therefore the input vector to the MLP has twice the node-feature's dimensions, since the direct sum ${\text{h}}_i^{(l)}\oplus ({\text{h}}_j^{(l)}-{\text{h}}_i^{(l)})$, is a concatenation of the two vector quantities of equal dimensions. The dimension of the MLP's output is the same as the output dimension of the message passing operations and has a linear activation. We fix the MLP to have two hidden layers with ReLU activation and the same number of nodes as the output dimension. The network is implemented using the Pytorch-Geometric (Fey and Lenssen, 2019) package. Note that we have not performed any hyperparameter scan as part of this present, proof-of-concept study. We train the network for fifty epochs with a learning rate of 0.001 using the Adam (Kingma and Ba, 2014) optimiser. The training and validation losses are compared after each epoch to ensure that there is no overfitting or a premature termination of training. The epoch with minimum validation loss is used to infer the anomaly detection on the test dataset.

4. Anomaly Detection Performance and Results

We now discuss the performance of the designed IRC safe loss function in detecting anomalous jets when the network is trained only on the QCD background. We choose boosted top jets from the aforementioned public dataset as our benchmark. In Figure 3 (left), we show the distribution of the loss function for the QCD and top jets (our inputs are the node features given in Equation 11). As can be seen, the distributions of the loss function values for the QCD and top jets are significantly different, highlighting the capability of the architecture to detect anomalous jets in an IRC-safe way. The Receiver-Operator-Characteristic (ROC) curve and the Area Under the Curve (AUC) of 0.902 shown in Figure 3 (right) confirm the good separation shown in the loss distribution, rivaling convolutional autoencoders (Heimel et al., 2019; Roy and Vijay, 2019; Farina et al., 2020; Finke et al., 2021) which also have AUCs close to such values [up to 0.93 (Heimel et al., 2019) and 0.91 (Finke et al., 2021)] on the same dataset. Although we did not perform a hyperparameter scan for this study, we observed a decrease in performance for a one-dimensional latent space.

FIGURE 3

Figure 3. The distribution of the loss function of an IRC safe graph autoencoder trained only with QCD jets with graph radius R₀ = 0.3.

Top jets possess a different and hard kinematical structure that is typically not present in QCD jets. The ability to look into the soft and collinear splittings from the QCD shower evolution in an IRC safe way enables the network to access such information and the hard radiation pattern in a theoretically meaningful way. Modifications of the soft and collinear radiation patterns that are seeded by novel hard scales (see, e.g., Englert et al., 2011; Soper and Spannowsky, 2011, 2014; Gerwick et al., 2012a,b; Prestel and Spannowsky, 2019 for a more traditional jet-based approach to this) are therefore consistently included in the anomaly detection performance. Therefore, when such non-QCD structures are present, the anomaly detection performance is considerably improved.

In light of these results, it is worthwhile to compare our autoencoder results to phenomenological aspects of QCD in jet substructure analyses. From the point of view of soft and collinear features, Energy Correlation Functions (ECF) (Larkoski et al., 2013) are particularly relevant for such a comparison as we will motivate below. Furthermore, given that our autoencoder condenses the QCD information into the latent space in an IRC-safe way, it is interesting to see how it correlates with ECF observables. To this end, we define

$\begin{array}{ll}\text{g}={\displaystyle \sum _{a\in {G}}}{z}_a{\text{h}}_a,& (12)\end{array}$

where h_a are the latent node features. Similar to the graph readout in a classification scenario (Konar et al., 2022), this is an IRC safe representation of the jet. The distribution of the individual components of the two-dimensional graph representation are shown in Figure 4. The good performance of the autoencoder is reflected in the good separation in the latent space. The two latent space directions are, however, completely anti-correlated; see Figure 5 (they are also highly correlated with the loss). Thus, restricting ourselves to any of these three variables would be sufficient for the anomaly detection problem studied in this work. The loss would most likely be a better choice when one focuses on anomaly detection capabilities since it condenses the information of the two-dimensional latent space into a single quantity. On the other hand, any latent feature would be more suitable for applications demanding lower execution times, since in this case only the encoder needs to be evaluated during inference.

FIGURE 4

Figure 4. The distribution of each dimension of the two-dimensional latent spaces obtained after an IRC safe graph readout given in Equation (12).

FIGURE 5

Figure 5. The correlation of IRC safe loss (cf. Equation 8) and latent dimension (obtained with Equation 12) is shown with the Energy Correlation Functions (13). One can see a very high correlation of the ECFs with the variables obtained from the network, hinting at a close connection between them.

Moving on to the relation of the learned information with ECFs, we first define the ECFs as

$\begin{array}{ll}\text{ECF}(N,\beta )={\displaystyle \sum _{i_1<i_2<....i_{N-1}<i_N\in J}}\left({\displaystyle \prod _{a=1}^N}{z}_{i_a}\right)\left({\displaystyle \prod _{b=1}^{N-1}}{\displaystyle \prod _{c=b+1}^N}\Delta R_{i_b{i}_c}^{\beta }\right).& (13)\end{array}$

Focussing specifically on the case N = 2, we obtain

$\begin{array}{ll}\text{ECF}(2,\beta )={\displaystyle \sum _{j<i\in I_J}}{z}_i{z}_j\Delta R_{ij}^{\beta }={\displaystyle \sum _{i=1}^{\left|J\right|}}{z}_i{\displaystyle \sum _{j=i+1}^{\left|J\right|}}{z}_j\Delta R_{ij}^{\beta },& (14)\end{array}$

where I_J is the index set of the constituent set J, and $z_i=p_T^i/({\sum }_{k\in I_J}{p}_T^k)$. We can rewrite the expression as

$\begin{array}{ll}\text{ECF}(2,\beta )={\displaystyle \sum _{i=1}^{\left|J\right|}}{z}_i{H}_i,H_i={\displaystyle \sum _{j=i+1}^{\left|J\right|}}{z}_j\Delta R_{ij}^{\beta }.& (15)\end{array}$

Therefore, the quantity H_i can be regarded as a scalar node feature obtained from the message function $\Delta R_{ij}^{\beta }=\Delta R_{ij}^{\beta }({\widehat{p}}_i,{\widehat{p}}_j)$, with a weighted (by z_j) summed readout, while the sum over i to get the ECF is similar to a graph readout operation on all the nodes (or constituents) of the jet. Although the graph structure in the current case is the 2-combinatorial graph, such an analogy suggests that the features extracted by the EMPN are closely connected to ECFs.

This expectation is analyzed in more detail in Figure 5, where we show the correlation of different order ECFs with each dimension g_i of the latent graph readout and the loss function. There is a strong correlation between the 2-point ECFs and the network outputs, which decreases when considering the 4-point ECFs. This difference illustrates the close relation of the message passing architecture to the 2-point ECFs. The latent dimensions show a higher correlation for β = 1 than β = 2, while the opposite holds for the loss function. This may be due to the ReLU activation, which is essentially a linear function for all positive arguments, while the loss function's higher correlation to the quadratic ECFs may be due to the usage of the mean-squared error as $d({\widehat{p}}_i,{\widehat{\stackrel{-}{p}}}_i)$.

5. Conclusions

Infrared and collinear safety is not a luxury but an essential requirement to guarantee the theoretical consistency of particle physics collider data interpretations. The emerging and fast-developing area of anomaly detection should therefore incorporate IRC safety when analyzing data at the LHC where QCD activity plays a dominant role. New heavy physics significantly deviates from QCD phenomenology, predominantly characterized by soft and collinear emissions. Reflecting the QCD expectation adequately helps isolate anomalies further; the ability to meaningfully interpolate into the soft and collinear regime is crucial for extending the reach of such techniques to lower scales. Despite this, IRC safety has not played an essential role in the implementation of anomaly detection. In this paper, we have placed IRC safety at the heart of anomaly detection for the first time by constructing a graph neural network autoencoder that employs Energy-Weighted Message-Passing, which gives rise to an IRC-safe architecture (Konar et al., 2022).

Graph neural networks are well-adapted approaches for isolating tell-tale correlations of final states (Atkinson et al., 2021a; Dreyer and Qu, 2021) and we find that our algorithm shows a high anomaly detection capability whilst having theoretically appealing properties. We have demonstrated this by injecting top jets as an anomaly and finding excellent discriminating sensitivity. While this partly results from the direct presence of a novel hard scale in the jet's substructure, additional sensitivity is accessed from a different soft and collinear shower pattern that accompanies the hard scale. To highlight this relation to well-studied observables in QCD phenomenology, we have shown a strong relation of the information encoded in our autoencoder's latent space with energy correlation functions. This motivates extending anomaly detection analyses using our framework to new physics scenarios of lighter BSM degrees of freedom, which we leave for future work.

Data Availability Statement

Publicly available datasets were analyzed in this study. This data can be found here: https://zenodo.org/record/2603256.

Author Contributions

OA, AB, and VN implemented the code and dataset used in the analysis. AB, CE, and VN wrote the first draft of the manuscript. All authors contributed to the conception and design of building the IRC safe graph autoencoder and contributed to manuscript revision, read, and approved the submitted version.

Funding

OA was supported by the UK Science and Technology Facilities Council (STFC) under Grant ST/V506692/1. AB and CE were supported by the STFC under Grant ST/T000945/1. CE was also supported by the Leverhulme Trust under Research Project Grant RPG-2021-031 and the IPPP Associate Scheme. PK and VN were supported by the Physical Research Laboratory (PRL), Department of Space, Government of India. MS was supported by the STFC under Grant ST/P001246/1. Part of the computational work detailed in this paper was performed using the HPC resources (Vikram-100 HPC) and TDP project at PRL.

Conflict of Interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The handling editor CD is currently organizing a Research Topic with the authors MS.

Publisher's Note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

1. ^This is not the case for popular graph construction algorithms like k-nearest neighbors, for which the decision and threshold has a complicated dependence on the distance of the primary node i with every other particle in the graph, and on the number of elements in the neighborhood set.

2. ^This is also the reason for using closed neighborhoods ${N}\left[i\right]$, as an open neighborhood ${N}(i)$, would give a special status to the node i.

3. ^The effective number of inputs to a message function could be twice the number of input features–one each for the two nodes connected by an edge. However, a concrete understanding of the universal approximation properties of graph neural networks (Brüel Gabrielsson, 2020) is yet to be achieved, making it difficult to precisely determine the actual input dimensions when looking at the complete graph neural network.

References

Aaboud, M., Aad, G., Abbott, B., Abdinov, O., Abeloos, B., Abidi, S. H., et al. (2019). A strategy for a general search for new phenomena using data-derived signal regions and its application within the ATLAS experiment. Eur. Phys. J. C 79, 120.

Google Scholar

Araz, J. Y., and Spannowsky, M. (2021). Combine and conquer: event reconstruction with Bayesian ensemble neural networks. J. High Energy Phys. 4, 296. doi: 10.1007/04(2021)296

CrossRef Full Text | Google Scholar

Atkinson, O., Bhardwaj, A., Brown, S., Englert, C., Miller, D. J., and Stylianou, P. (2021a). Improved constraints on effective top quark interactions using edge convolution networks. J. High Energy Phys. 2022:137. doi: 10.1007/04(2022)137

CrossRef Full Text | Google Scholar

Atkinson, O., Bhardwaj, A., Englert, C., Ngairangbam, V. S., and Spannowsky, M. (2021b). Anomaly detection with convolutional graph neural networks. J. High Energy Phys. 8, 80. doi: 10.1007/08(2021)080

CrossRef Full Text | Google Scholar

Bakshi, S. D., Chakrabortty, J., Englert, C., Spannowsky, M., and Stylianou, P. (2022). Landscaping CP-violating BSM scenarios. Nucl. Phys. B 975, 115676. doi: 10.1016/j.nuclphysb.2022.115676

CrossRef Full Text | Google Scholar

Blance, A., and Spannowsky, M. (2020). Unsupervised event classification with graphs on classical and photonic quantum computers. J. High Energy Phys. 21, 170. doi: 10.1007/08(2021)170

CrossRef Full Text | Google Scholar

Blance, A., Spannowsky, M., and Waite, P. (2019). Adversarially-trained autoencoders for robust unsupervised new physics searches. J. High Energy Phys. 10, 047. doi: 10.1007/10(2019)047

CrossRef Full Text | Google Scholar

Brüel Gabrielsson, R. (2020). “Universal function approximation on graphs,” in Advances in Neural Information Processing Systems, 33, eds H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, and H. Lin (Curran Associates, Inc.), 19762–19772.

Google Scholar

Butter, A., Kasieczka, G., Plehn, T., and Russell, M. (2018). Deep-learned top tagging with a Lorentz layer. SciPost Phys. 5, 28. doi: 10.21468/SciPostPhys.5.3.028

CrossRef Full Text

Cacciari, M., Salam, G. P., and Soyez, G. (2008). The anti-k_t jet clustering algorithm. J. High Energy Phys. 4, 63. doi: 10.1088/1126-6708/2008/04/063

CrossRef Full Text

Canelli, F., de Cosa, A., Pottier, L. L., Niedziela, J., Pedro, K., and Pierini, M. (2022). Autoencoders for semivisible jet detection. J. High Energy Phys. 2, 74. doi: 10.1007/02(2022)074

CrossRef Full Text | Google Scholar

Carmona, A., Lazopoulos, A., Olgoso, P., and Santiago, J. (2021). Matchmakereft: automated tree-level and one-loop matching. SciPost Phys. 12, 198. doi: 10.21468/SciPostPhys.12.6.198

CrossRef Full Text | Google Scholar

Catani, S., Dokshitzer, Y. L., Seymour, M. H., and Webber, B. R. (1993). Longitudinally invariant K_t clustering algorithms for Hadron Hadron collisions. Nuclear Phys. B 406, 187–224. doi: 10.1016/0550-3213(93)90166-M

CrossRef Full Text | Google Scholar

Charles, R. Q., Su, H., Kaichun, M., and Guibas, L. J. (2017). “PointNet: deep learning on point sets for 3D classification and segmentation,” in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), (Honolulu, HI: IEEE Computer Society), 77–85. doi: 10.1109/CVPR.2017.16

CrossRef Full Text | Google Scholar

Cheng, T., Arguin, J.-F., Leissner-Martin, J., Pilette, J., and Golling, T. (2020). Variational autoencoders for anomalous jet tagging. arXiv preprint arXiv:2007.01850. doi: 10.48550/arXiv.2007.01850

CrossRef Full Text | Google Scholar

Collins, J. C., Soper, D. E., and Sterman, G. F. (1989). Factorization of hard processes in QCD. Adv. Ser. Direct. High Energy Phys. 5, 1–91. doi: 10.1142/9789814503266_0001

CrossRef Full Text | Google Scholar

Collins, J. H., Howe, K., and Nachman, B. (2018). Anomaly detection for resonant new physics with machine learning. Phys. Rev. Lett. 121, 241803. doi: 10.1103/PhysRevLett.121.241803

PubMed Abstract | CrossRef Full Text | Google Scholar

Collins, J. H., Martin-Ramiro, P., Nachman, B., and Shih, D. (2021). Comparing weak- and unsupervised methods for resonant anomaly detection. Eur. Phys. J. 81:617. doi: 10.1140/epjc/s10052-021-09389-x

CrossRef Full Text | Google Scholar

Das Bakshi, S., Chakrabortty, J., Englert, C., Spannowsky, M., and Stylianou, P. (2021). CP violation at ATLAS in effective field theory. Phys. Rev. D 103, 055008. doi: 10.1103/PhysRevD.103.055008

PubMed Abstract | CrossRef Full Text | Google Scholar

de Favereau, J., Delaere, C., Demin, P., Giammanco, A., Lemaetre, V., Mertens, A., et al. (2014). DELPHES 3, a modular framework for fast simulation of a generic collider experiment. J. High Energy Phys. 2, 57. doi: 10.1007/02(2014)057

CrossRef Full Text

De Simone, A., and Jacques, T. (2019). Guiding new physics searches with unsupervised learning. Eur. Phys. J. C 79, 289. doi: 10.1140/epjc/s10052-019-6787-3

CrossRef Full Text | Google Scholar

Dillon, B. M., Plehn, T., Sauer, C., and Sorrenson, P. (2021). Better latent spaces for better autoencoders. SciPost Phys. 11, 61. doi: 10.21468/SciPostPhys.11.3.061

CrossRef Full Text | Google Scholar

Dolan, M. J., and Ore, A. (2021). Equivariant energy flow networks for jet tagging. Phys. Rev. D 103, 074022. doi: 10.1103/PhysRevD.103.074022

CrossRef Full Text | Google Scholar

Dreyer, F. A., and Qu, H. (2021). Jet tagging in the Lund plane with graph networks. J. High Energy Phys. 3, 52. doi: 10.1007/03(2021)052

CrossRef Full Text | Google Scholar

Englert, C., Galler, P., Pilkington, A., and Spannowsky, M. (2019). Approaching robust EFT limits for CP-violation in the Higgs sector. Phys. Rev. D 99, 095007. doi: 10.1103/PhysRevD.99.095007

CrossRef Full Text | Google Scholar

Englert, C., Galler, P., and White, C. D. (2020). Effective field theory and scalar extensions of the top quark sector. Phys. Rev. D 101, 035035. doi: 10.1103/PhysRevD.101.035035

CrossRef Full Text | Google Scholar

Englert, C., Plehn, T., Schichtel, P., and Schumann, S. (2011). Jets plus missing energy with an autofocus. Phys. Rev. D 83, 095009. doi: 10.1103/PhysRevD.83.095009

CrossRef Full Text | Google Scholar

Farina, M., Nakai, Y., and Shih, D. (2020). Searching for new physics with deep autoencoders. Phys. Rev. D 101, 075021. doi: 10.1103/PhysRevD.101.075021

CrossRef Full Text | Google Scholar

Fey, M., and Lenssen, J. E. (2019). “Fast graph representation learning with PyTorch Geometric,” in ICLR Workshop on Representation Learning on Graphs and Manifolds (New Orleans, LA).

Google Scholar

Finke, T., Krämer, M., Morandini, A., Mück, A., and Oleksiyuk, I. (2021). Autoencoders for unsupervised anomaly detection in high energy physics. J. High Energy Phys. 6, 161. doi: 10.1007/06(2021)161

PubMed Abstract | CrossRef Full Text | Google Scholar

Freitas, A., López-Val, D., and Plehn, T. (2016). When matching matters: loop effects in Higgs effective theory. Phys. Rev. D 94, 095007. doi: 10.1103/PhysRevD.94.095007

CrossRef Full Text | Google Scholar

Gerwick, E., Plehn, T., and Schumann, S. (2012a). Understanding jet scaling and jet vetos in Higgs searches. Phys. Rev. Lett. 108, 032003. doi: 10.1103/PhysRevLett.108.032003

PubMed Abstract | CrossRef Full Text | Google Scholar

Gerwick, E., Plehn, T., Schumann, S., and Schichtel, P. (2012b). Scaling patterns for QCD jets. J. High Energy Phys. 10, 162. doi: 10.1007/10(2012)162

PubMed Abstract | CrossRef Full Text | Google Scholar

Hajer, J., Li, Y.-Y., Liu, T., and Wang, H. (2020). Novelty detection meets collider physics. Phys. Rev. D 101, 076015. doi: 10.1103/PhysRevD.101.076015

CrossRef Full Text | Google Scholar

Hallin, A., Isaacson, J., Kasieczka, G., Krause, C., Nachman, B., Quadfasel, T., et al. (2021). Classifying Anomalies THrough Outer Density Estimation (CATHODE). arXiv preprint arXiv:2109.00546. doi: 10.48550/arXiv.2109.00546

CrossRef Full Text | Google Scholar

Heimel, T., Kasieczka, G., Plehn, T., and Thompson, J. M. (2019). QCD or what? SciPost Phys. 6, 30. doi: 10.21468/SciPostPhys.6.3.030

CrossRef Full Text | Google Scholar

Kasieczka, G., Plehn, T., Thompson, J., and Russel, M. (2019). Top quark tagging reference dataset. Available online at: https://zenodo.org/record/2603256#.YsWdc9JBwsc

Kilgore, W. B., and Giele, W. T. (1997). Next-to-leading order gluonic three jet production at hadron colliders. Phys. Rev. D 55, 7183–7190. doi: 10.1103/PhysRevD.55.7183

CrossRef Full Text | Google Scholar

Kingma, D. P., and Ba, J. (2014). Adam: a method for stochastic optimization. arXiv preprint arXiv:1412.6980v9. doi: 10.48550/arXiv.1412.6980

CrossRef Full Text | Google Scholar

Kinoshita, T. (1962). Mass singularities of Feynman amplitudes. J. Math. Phys. 3, 650–677. doi: 10.1063/1.1724268

CrossRef Full Text | Google Scholar

Komiske, P. T., Metodiev, E. M., and Thaler, J. (2019). Energy flow networks: deep sets for particle jets. J. High Energy Phys. 1, 121. doi: 10.1007/01(2019)121

CrossRef Full Text | Google Scholar

Konar, P., Ngairangbam, V. S., and Spannowsky, M. (2022). Energy-weighted message passing: an infra-red and collinear safe graph neural network algorithm. J. High Energy Phys. 2, 60. doi: 10.1007/02(2022)060

CrossRef Full Text | Google Scholar

Larkoski, A. J., Salam, G. P., and Thaler, J. (2013). Energy correlation functions for jet substructure. J. High Energy Phys. 6, 108. doi: 10.1007/06(2013)108

PubMed Abstract | CrossRef Full Text | Google Scholar

Lee, T. D., and Nauenberg, M. (1964). Degenerate systems and mass singularities. Phys. Rev. 133, B1549–B1562. doi: 10.1103/PhysRev.133.B1549

CrossRef Full Text | Google Scholar

Nachman, B. (2020). Anomaly detection for physics analysis and less than supervised learning. arXiv preprint arXiv:2010.14554. doi: 10.48550/arXiv.2010.14554

CrossRef Full Text | Google Scholar

Nachman, B., and Shih, D. (2020). Anomaly detection with density estimation. Phys. Rev. D 101, 075042. doi: 10.1103/PhysRevD.101.075042

CrossRef Full Text | Google Scholar

Prestel, S., and Spannowsky, M. (2019). HYTREES: combining matrix elements and parton shower for hypothesis testing. Eur. Phys. J. C 79, 546. doi: 10.1140/epjc/s10052-019-7030-y

CrossRef Full Text | Google Scholar

Qi, C. R., Yi, L., Su, H., and Guibas, L. J. (2017). “PointNet++: deep hierarchical feature learning on point sets in a metric space,” in Advances in Neural Information Processing Systems, eds I. Guyon, U. Von Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwananthan, and R. Garnett (Long Beach, CA), 30.

Google Scholar

Roy, T. S., and Vijay, A. H. (2019). A robust anomaly finder based on autoencoders. arXiv preprint arXiv:1903.02032. doi: 10.48550/arXiv.1903.02032

CrossRef Full Text | Google Scholar

Sjöstrand, T., Ask, S., Christiansen, J. R., Corke, R., Desai, N., Ilten, P., et al. (2015). An introduction to PYTHIA 8.2. Comput. Phys. Commun. 191, 159–177. doi: 10.1016/j.cpc.2015.01.024

CrossRef Full Text | Google Scholar

Sjostrand, T., Mrenna, S., and Skands, P. Z. (2008). A Brief Introduction to PYTHIA 8.1. Comput. Phys. Commun. 178, 852–867. doi: 10.1016/j.cpc.2008.01.036

CrossRef Full Text | Google Scholar

Soper, D. E., and Spannowsky, M. (2011). Finding physics signals with shower deconstruction. Phys. Rev. D 84, 074002. doi: 10.1103/PhysRevD.84.074002

CrossRef Full Text | Google Scholar

Soper, D. E., and Spannowsky, M. (2014). Finding physics signals with event deconstruction. Phys. Rev. D 89, 094005. doi: 10.1103/PhysRevD.89.094005

CrossRef Full Text | Google Scholar

Wang, Y., Sun, Y., Liu, Z., Sarma, S. E., Bronstein, M. M., and Solomon, J. M. (2019). Dynamic graph CNN for learning on point clouds. ACM Trans. Graph. 38, 1–12. doi: 10.1145/3326362

CrossRef Full Text | Google Scholar

Weinberg, S. (1979). Phenomenological Lagrangians. Phys. A 96, 327–340. doi: 10.1016/0378-4371(79)90223-1

CrossRef Full Text | Google Scholar

Zaheer, M., Kottur, S., Ravanbakhsh, S., Poczos, B., Salakhutdinov, R. R., and Smola, A. J. (2017). “Deep sets,” in Advances in Neural Information Processing Systems (La Jolla, CA), 30.